- Works
- 370
- Views
- 806
- Likes
- 29
- Updated
Kevin Abosch
By early 2015, neural networks had mastered the art of 'image-to-text' and could create natural language captions for images. Flipping this process, and turning text into image, was a much more complex challenge solved by 19-year old prodigy Elman Mansimov's alignDRAW model.Fellowship is pleased to present a special release of fully on-chain* NFTs of this historical artwork, containing all the original 32x32 pixel images created in 2015.
Elman Mansimov
A group of happy elephants in the green grass field #103
An airplane with its landing wheels out landing #75
A blue school bus parked in a parking lot #45
A green school bus parked in a parking lot #63
A yellow school bus parked in a parking lot #57
A picture of a dark sky #27
A yellow school bus is flying in blue skies #3
A yellow school bus is walking across a green grass field #9
Sofia Crespo
A person skiing on sand clad vast desert #1
A very large commercial airplane walking in the green grass field #53
A toilet seat sits open in an empty bathroom #16
[[[vessel_within_vessel_0099]]]
A person skiing on snow clad vast mountain #81
A picture of a morning sky #41
An airplane flying off into the distance at night #116
A very large commercial airplane sitting on surfboard in the water #9
A group of happy elephants in the dry grass field #23
A brown horse is grazing in a beach #44
An airplane flying off into the distance on a clear day #83
A brown horse is grazing in a field #73
A herd of elephants flying in the blue skies #105
A toilet seat sits open in the grass field #115
“Crossing the Interface (DAO),” it adapts a 2014 performance at the Guggenheim Museum in New York. Each piece in the series is a short, AI-generated video that draws on documentation of the earlier work. The animations project the worried, meandering line characteristic of GAN art into the temporal dimension. The visual field is organized by shapes that move in repetitive, quasi-recursive patterns, creating a feeling of both movement and stasis.
Crossing the Interface (DAO) V
Mario Klingemann (Quasimondo)
Memo Akten
An exploration into the minds of an artificial, and a biological, neural network. Continuing on from my "Journey through the layers of the mind" (2015), "Learning to See" (2017-) and "Deep Meditations" series (2018-); continuing investigations into Deep Visual Instruments for Latent Story-telling - Deep Learning models as an artistic medium for new modes of performative, creative expression with abstract narrative.
Created using custom software using VQGAN+CLIP.
Roope Rainisto
Anne Spalter
A spaceship flying through sunrise, day, sunset, and night. Created with the help of text-to-image AI (VQGAN+CLIP).
TODEM by Jason Salavon
TODEM (Tapestry of Decadent Meritocracy) by Jason Salavon is a vast AI-generated digital tapestry, composed of 1000 unique tiles, forming an animated GIF with dimensions of 100K x 58K pixels—potentially one of the largest animations ever created. Each tile is a standalone NFT, minted on Ethereum, and reflects themes of wealth distribution and meritocracy, with pricing based on pixel area to critique inequality.
Using tools like Stable Diffusion, Deforum, and GPT-3.5, and datasets from Wikipedia, FFHQ, and the Federal Reserve Bank of St. Louis, Salavon transforms text prompts into intricate, quasi-recursive animations. The work blends humor and discomfort, highlighting AI’s amoral yet creative potential. Part of the proceeds supports structural reforms like rank-choice voting.
TODEM merges conceptual depth with technical innovation, challenging perceptions of AI’s role in art and its implications for society.
See the full GIF [here] (https://latentculture.com/todem/).
Ivona Tau collaborated with Tiga James Sontag to create visuals for a track celebrating Unified Dancefloor Humanity (UDH). Using her archive of concert photography spanning nearly a decade, she trained a custom AI model with thousands of curated images capturing diverse musical performances and audiences united by their love for music. Through generative adversarial networks (GAN) and abstract imagery, she crafted a universal representation of the collective energy and passion of crowds inspired by sound, transcending boundaries of style, genre, and location.
Spaceship: Full Day Flight
The Service of Silence
Deeper Meditations #1 (1/1)
Contortion
Mike Tyka
"Portraits of Imaginary People" by Mike Tyka is a groundbreaking exploration of identity and authenticity through the lens of artificial intelligence. Using StackGAN (Stacked Generative Adversarial Networks), Tyka creates hyper-realistic portraits of individuals who do not exist, blurring the boundaries between reality and fiction. This work draws inspiration from digital disinformation phenomena, particularly fabricated identities used in online propaganda, such as the Russian troll accounts that impersonated real individuals during the 2016 U.S. elections. Through these AI-generated faces, Tyka raises critical questions about the ethics of digital fabrication and the implications of technology in shaping perception and trust.
Technically, the project employs a multi-stage GAN architecture to generate high-resolution portraits. Starting with a low-resolution image (128x128 or 256x256 pixels), successive stages progressively upscale and refine the details, achieving resolutions up to 4k x 4k. This layered approach enables intricate textures and vivid realism, transforming simple pixel patterns into lifelike representations. Despite aiming for high realism, Tyka embraces artistic imperfections and emergent styles, allowing for a fusion of realism and surrealism in the final outputs.
paolakinck
MaryCarry92
Refik Anadol’s Synthetic Dreams – Landscapes represents a groundbreaking fusion of quantum computing and generative art, created using a Generative Adversarial Network (GAN) in conjunction with quantum data. This series is part of Anadol’s exploration of Quantum AI Data Paintings and leverages quantum bit strings generated by the Google Quantum AI team’s “beyond classical” experiment. These quantum inputs, marked by their blend of randomness and precision, were processed through a GAN algorithm, likely StyleGAN2, to produce a series of mesmerizing digital landscapes.
The artworks draw from a dataset of 200 million global landscape images, including visuals from all U.S. national parks. Using custom noise algorithms and quantum measurements, Anadol guided the generative process to synthesize images that combine the realism of natural landscapes with the abstract beauty of quantum randomness. Each of the 1,000 unique digital artworks in the series is computed using a distinct quantum bit string, resulting in a rich interplay between organic patterns and the poetic sublimity of Earth’s textures.
The resulting works blend earthy pigments, intricate patterns, and structural coherence, evoking a sensory connection to the natural world. By merging quantum data with cutting-edge GAN techniques, Synthetic Dreams – Landscapes bridges art, technology, and science, offering a poetic exploration of human perception, nature, and innovation. This series not only reimagines the essence of Earth’s landscapes but also pushes the boundaries of AI-driven generative art and quantum-inspired creativity.
aurèce vettier (Paul Mouginot)
Potential Herbariums is a groundbreaking project by Aurèce Vettier (Paul Mouginot), blending generative algorithms with traditional botanical aesthetics to explore the intersection of nature, technology, and art. This series reimagines botanical forms as digital artifacts, presenting a speculative archive of plant species that exist only in the realm of artificial intelligence.
The project employs AI models trained on a curated dataset of botanical illustrations, scientific diagrams, and natural imagery. By synthesizing these inputs, the algorithm generates intricate and imaginative plant-like structures, each imbued with a sense of organic authenticity yet unmistakably artificial. These digital herbariums challenge conventional understandings of biodiversity and push the boundaries of how we perceive and document the natural world.
Through Potential Herbariums, Aurèce Vettier reflects on themes of ecological memory, the fragility of the natural environment, and the potential for AI to serve as both a mirror and an extension of human creativity. The series also underscores the interplay between historical botanical archives and contemporary digital technologies, offering a poetic meditation on the evolving relationship between humanity, nature, and machine intelligence.
element/pass1fl0ra
Bård Ionson
Drone Wars by Bard Ionson explores the dystopian implications of automation and artificial intelligence in modern warfare. Using cGAN (Conditional GAN) and pix2pix, the artwork transforms abstract inputs into haunting representations of drones, propaganda, and automated conflict. Ionson’s use of pix2pix enables a dynamic interplay between reality and machine-generated abstraction, reflecting the technological detachment of war from human agency.
Thematically, Drone Wars critiques the cyclical nature of propaganda and the flawed assumption that the future can be programmed based on the past. The AI-generated imagery, conditioned by Ionson’s input, evokes distorted, surreal visions of warfare, highlighting the ethical dilemmas of machine-led systems. The accompanying text amplifies these concerns, questioning humanity’s reliance on automation to shape critical decisions.
Through its innovative use of generative AI and its philosophical depth, Drone Wars serves as a powerful commentary on the intersection of technology, power, and the erosion of accountability in the age of automated conflict.
RE: ROAD TRIP by Margaret Murphy is a series of 200 AI-generated images reimagining the historically influential photographic genre of the road trip through the perspective of the contemporary female gaze. Murphy considers how this revered category has shaped and defined contemporary fine art photography using style influences, including artists such as Stephen Shore, Ed Ruscha, and Joel Sternfeld.
daily.xyz AIartwork - daily curated
emprops.ai
Day #4 - Memory
Alsoguppyme: Day #17 - Body parts
Day #11 - We Just Got Here
Running a Fever: Day #7 - Hurricane
"Taking 'The Muses' project as a whole—as the entire long-form collection, not just the individual works—and assuming that the "male gaze", the "female gaze", and the "male gaze reproduced by women" are all mashed up, almost remixed, across its 500 images, what does that tell us about the collection's overall meaning?
With artificial intelligence's help, can "The Muses" point the way to constructing a new type of "gaze", removed from gender assignment?" (Anika Meier)
"The Myth of Dream" by Marlon Hacla is a pioneering work that explores the intersection of artificial intelligence, visual storytelling, and human imagination. In this series, Hacla trained an AI model using scanned illustrations from children’s books, allowing the machine to generate dreamlike, surreal images that reinterpret and expand upon traditional art forms.
Each piece in the collection reflects the AI’s unique capacity to blend patterns, colors, and forms into compositions that evoke the fragmented, fluid nature of dreams. The work challenges viewers to reconsider the boundaries between human and machine creativity, highlighting how AI can transform familiar narratives into entirely new visual experiences.
Hacla’s use of AI not only reimagines the aesthetics of storytelling but also invites reflection on the role of technology in shaping contemporary art. As part of this innovative series, the artwork embodies a fusion of tradition and futurism, making it a standout example of the potential for collaboration between artists and artificial intelligence.
Vortex of dreams
Depicting dreams has long been a practice that borders on something almost absurd. They are intricate, abstract, fragmented and mostly fleeting visuals manifested only in one person, and when shared with others they seem to take a step toward a more grounded existence. Perhaps this grounding of the ephemeral emphasises the intimacy of sharing dreams and, of course, makes them somewhat more real.
In le travail des rêves, aurèce vettier does exactly this. He invites the viewer to participate in his personal dreamscape, encouraging us to submerge ourselves in his memories, from childhood until current times. Though with an indication in the prompt used for its creation and in the title of each piece, the viewer is still left with space for interpretation. Furthermore, with the low resolution of the pieces, the viewer is forced to search their own memories and dreamscape and, perhaps, also reflect upon what dreams might symbolise.
Agoria (Sébastien Devaud)
Yuma Kishi (Obake AI)
loopytezoverse
Auntie's Spa: Legitimate Business
Alice Gordon
Irina Angles and Dr Formalyst
Almost Human
Fragile Memories #05: Loneliness, On An Autumn Evening
The Agoraphobic Traveller (Jacqui Kenny)
Alexander Mordvintsev
Eponym
fairy tale
Helena Sarin (NeuralBricolage)
Disproportionate Anxiety
Cognitiv Behaviour
Self-awareness
Third Eye Implant
Day #35 - Black Sunday
The Age of Self Indulgence
Journey through the layers of the mind
1111
Sun Signals
888
COMMENT OUT
Temporary Permanence
Excessize
Los Angelizing delves into the exaggerated cultural perceptions of Los Angeles, shedding light on the city's hidden complexities. Through the use of AI, Murphy amplifies depictions of the subtle extremes of the city’s iconic imagery, challenging viewers to reconsider their preconceptions of the city’s identity and its inhabitants.
viña del mar
Margaret Murphy
RE: ROAD TRIP
Los Angelizing
Oscar After Party
a perfect storm
Generation 2 (EpoStory)
Generation 1
Running a Fever Day
X New Worlds
Dancevatar
nouseskou
(loop (format t "~%"))
Osf 2
Oss 1
Pierre Zandrowicz
Alsoguppyme
pygmalion
Vacation
Slumber Party
Visit The Cave
Flower Electric
Barn Quilt Show
AI Spaceships
Outerspace Steampunk
Outerspace Interceptor
Outerspace Sentient
Lumina - The Convergence of Death
chaotic serenity
tiananmen square massacre
alignDRAW
A Father And His Child In Front Of A Lighthouse
DreamScapes
Xander Steenbrugge
Harvest
Golan Levin
These faces are “ambigrams”: images that are legible both upside-down and right-side up. Created with a machine learning system, they may be displayed in any orientation. In this project, a collection of 55 such ambigrammatic faces have been generated in high resolution.
In the eighteenth and nineteenth centuries, bivalent face illusions were often used to depict uncomplicated dualities, such as young-old, good-evil, or blessed-damned. The faces in the Ambigrammatic Figures deck reflect the moral ambiguities of a darker and more uncertain time, marked by ecological crisis, misinformation, identitarianism, patriarchal authoritarianism, and the social unrest of a polity divided against itself.
Portraits of Imaginary People
Ambigrammatic Figures
Potential Herbarium
Seeds
Generating Images from Captions with Attention by Elman Mansimov, Emilio Parisotto, Jimmy Lei Ba & Ruslan Salakhutdinov [Submitted on 9 Nov 2015 (v1), last revised 29 Feb 2016 (this version, v2)]
"Motivated by the recent progress in generative models, we introduce a model that generates images from natural language descriptions. The proposed model iteratively draws patches on a canvas, while attending to the relevant words in the description. After training on Microsoft COCO, we compare our model with several baseline generative models on image generation and retrieval tasks. We demonstrate that our model produces higher quality samples than other approaches and generates images with novel scene compositions corresponding to previously unseen captions in the dataset [...]
In this paper, we illustrate how sequential deep learning techniques can be used to build a conditional probabilistic model over natural image space effectively. By extending the Deep Recurrent Attention Writer (DRAW) (Gregor et al., 2015), our model iteratively draws patches on a canvas, while attending to the relevant words in the description. Overall, the main contributions of this work are the following: we introduce a conditional alignDRAW model, a generative model of images from captions using a soft attention mechanism."
Conditional alignDRAW Model
DeepDream
GAN
Holly Herndon & Mat Dryhurst
Memo Akten
Seahorses Blooming in Abundance
"Seahorses Blooming in Abundance" (created with Stable Diffusion 1.5) The seafloor whispers with every grain of sand as creatures of myth glide through the water, a myriad of forms dancing in a silent ballet. Hues of coral and ocean currents blend, embracing in an aquatic mosaic. Seahorses, delicate and fierce, champions of the gentle depths, cradle life's fragile essence in their coiled tails. Their bodies carry the weight of beauty, slender and curving, echoing the motion of the waves above. Emerging life finds sanctuary in their form—so slender, so serene. The vast blue deep is laden with the echo of mystery, each creature a testament to the enigma of the marine world. Seahorses, emblems of fluidity and patience, are in a symphony of slow undulation. They conjure growth in stillness—the patient unfolding of life beneath the surface. Bubbles of possibility rise, drifting towards an unknowable destiny above. The motion is perpetual, ethereal, the seahorses' dance transcends the mere concept of time. Nature's ingenuity is revealed in each curve, in the stark ridges of their bodies, as selection's hand sculpts them in the quiet forge of evolution. What insights do these beings hold in their unhurried journey through the depths, as custodians of life's continuity? Their silent procession is a testament to the strength in softness, the power within vulnerability—a parable etched in living form.
The Awakening Of The Forest
Recurrent Neural Networks (RNN)
Neural Networks
VQGAN + CLIP
Convolutional Neural Network (CNN)
Transformer Model
Multimodal Transformers
Holly Herndon & Mat Dryhurst
Natural Language Processing (NLP): Large Language Models (LLM)
Infinite Images
Sasha Stiles
Spannungsbogen
Hibiscus
Linda Dounia
Homes of Argleton Lane
Ivona Tau
Rare Scrilla
Madonna
Entangled Others
Emergent Finale 00069
Chimerical Stories
Oxia Palus
DALL-E 2
Ixion
Cupid and Psyche
"Having the DALL-E 2 partially trained on the relevant artist’s surviving work, and then prompting it with textual input to guide its generation allows us to explore what a piece without visual ground truth might have been like. The reliability of the output then rests on two domains: work with which the artist produced during the same period as the lost piece, and textual accounts of its appearance. Whilst this will most likely not recreate the exact original piece, it can give us a range of possibilities as to what the original piece might have been like, and so offers a new strand of insight into art history, aided by machine learning." Continue reading
Diego Velázquez’s Cupid and Psyche was originally painted for the Hall of Mirrors in Madrid’s Royal Alcázar Palace but was tragically lost in the devastating fire of December 24, 1734. This painting, along with other masterpieces, is known only through historical records, leaving its visual essence shrouded in mystery for nearly 300 years.
Through the pioneering efforts of Oxia Palus, the lost artwork has been reimagined. Using advanced AI technology, including DALL·E 2, and extensive textual research, Oxia Palus has digitally reconstructed Cupid and Psyche, offering a compelling interpretation of Velázquez’s original masterpiece.
This reconstruction bridges the gap between history, art, and innovation, inviting viewers to experience a long-lost work through a fusion of cutting-edge technology and deep historical inquiry. It stands as a testament to how artificial intelligence can preserve and reimagine our cultural heritage.
Diego Velázquez’s Ixion was a depiction of the mythological figure from Greek legend, known for being bound to a fiery, eternal wheel as punishment for his transgressions. Painted for the Royal Alcázar Palace, it was part of a collection that celebrated mythological themes through Velázquez's masterful interpretation. Tragically, the painting was destroyed in the devastating fire of December 24, 1734, which consumed much of the palace's artwork.
Through meticulous textual research and advanced AI tools, Oxia Palus has reconstructed this lost work. Drawing on descriptions of the painting’s composition and Velázquez’s characteristic style, the reimagined Ixion captures the drama and emotional depth of the original while invoking the mythological gravitas that defined Velázquez's approach to classical subjects.
Beneath Leonardo da Vinci's Virgin of the Rocks lies the faint outline of an earlier composition: an alternative depiction of The Virgin and Christ Child set in a rocky landscape. This hidden sketch, revealed through x-ray imaging, provides a glimpse into da Vinci's creative process and the evolving vision for his masterpiece.
Oxia Palus used cutting-edge AI techniques to reconstruct this lost work. By training a GAN-based model on Leonardeschi paintings—artworks influenced by da Vinci—they mapped x-ray data of the hidden underdrawing to stylistically faithful reconstructions. The process involved co-registering x-ray images from Imperial College London and the National Gallery, where Virgin of the Rocks is housed, and augmenting these with variations in brightness, contrast, and sharpness.
The reconstructed Madonna transforms these augmented x-ray segments into vibrant paintings, reimagining the hidden artwork with remarkable fidelity. This project, presented at NVIDIA's 2020 GTC AI Art Gallery, highlights the convergence of art history and technology, offering a renewed perspective on da Vinci’s creative legacy.
Oxia Palus is an innovative project at the crossroads of art, technology, and cultural heritage, aiming to reconstruct lost or hidden artworks using advanced AI techniques. The project employs Conditional GANs (cGANs) like Pix2PixHD and Neural Style Transfer (CNN) to digitally restore ghost paintings uncovered through x-radiography and other imaging methods.
A notable achievement includes the reconstruction of Leonardo da Vinci’s “Madonna,” where x-ray imaging was combined with stylistic elements from related artworks to create a coherent visual representation. Similarly, hidden compositions in Pablo Picasso’s paintings have been revived, shedding light on the creative processes and decisions behind these masterpieces.
By blending cutting-edge machine learning with art historical expertise, Oxia Palus enhances the visualization of underdrawings and ghost paintings, providing color and stylistic consistency to what were previously grayscale images. This work not only broadens our understanding of artists’ oeuvres but also highlights the potential of human-AI collaboration in art conservation and analysis, preserving cultural heritage for future generations.
COMPOSE
MEMENTO MEMORIAE
Karma´s a glitch
element/dusk colorimeter
Sunday, 9 October 2022 11:09:26 CET (created with Stable Diffusion 1.5)
Points of Departure
Deeper Meditations
Deeper Meditations #6
Verses/Distributed Consciousness
Jon Wubbushi
Refik Anadol
Compend-AI-M
The Crisis
realTEN_GOP
hairy situation
Cellular Woman by Bård Ionson explores synthetic identity and machine perception by using StyleGAN2 to generate human faces that never existed. These faces are transformed into abstract representations inspired by the Excel spreadsheet art of Oleksiy Sai. The AI, instead of recognizing facial patterns, focuses on the grid-like structure of the spreadsheet, highlighting the biases and limitations of machine learning. By blending GAN-generated imagery with geometric abstraction, the artwork challenges the viewer to reflect on the fragility of AI interpretation and the artificiality of digital identity. Cellular Woman redefines generative art, merging technology and conceptual abstraction into a striking commentary on the relationship between humans and machines.
Cellular Woman
DISTANCE
Blue Hours
Ivona Tau’s "Blue Hours" series delves into the fleeting beauty of twilight, capturing the delicate interplay of light and shadow as the day transitions into night. Drawing on her personal photography, Tau trains neural networks to reinterpret these serene and introspective moments through the lens of AI. The resulting artworks blur the boundaries between reality and abstraction, blending the organic imperfections of photography with the generative creativity of machine learning.
Through this series, Tau explores the emotional resonance of liminal spaces—those in-between moments that evoke a sense of calm, introspection, and mystery. The compositions are painterly and dreamlike, with soft gradients and muted tones that mimic the quiet intensity of dusk. Each piece invites viewers to reflect on the transient nature of time and the interplay between human perception and artificial creativity.
"Blue Hours" exemplifies Tau’s ability to merge personal expression with technological innovation, offering a poetic meditation on the universal yet fleeting beauty of twilight.
Albertine Meunier
HyperChips
Diffusion Models
Fanny et Albertine
Anne et Albertine
Artnome et Albertine
Token Angel et Albertine
Nina et Albertine
Ana Maria et Albertine
Sasha et Albertine
Luluxxx et Albertine
Benôit et Albertine
Albertine Meunier
En Maillot de Bain
The series "En Maillot de Bain" by Albertine Meunier is a collection of AI-generated artworks created using StableDiffusion v2.1. The series explores the boundaries of AI's modesty and censorship capabilities by playing with the representation of bodies. Interestingly, the algorithm automatically blurs body details when the prompt is given in English ("in bathing suit"), but not when the same phrase is used in French ("en maillot de bain"). This concept of "algorithmic modesty" adds a playful and thought-provoking layer to the work. Source
Kevin Abosch
CIVICS
AI Mashups
AI Women
The Muses
Danielle King
Custom Machine Learning Algorithm
Marlon Hacla
Arena of the Gods
CIVICS by Kevin Abosch is an AI-generated NFT series featuring 100 unique artworks that explore themes of civil unrest, protest, and societal tensions. Each piece depicts surreal, photojournalistic scenes of activism in different global locations. Abosch used Stable Diffusion as the core technique to create these artworks, training the model on a mix of custom and public photographic datasets. This diffusion model allowed him to craft hyper-realistic yet distorted visuals, capturing the emotional weight and complexity of the events represented.
The use of Stable Diffusion enhances the surreal and manipulated aesthetics of CIVICS, blurring the line between documentary-style photography and synthetic art. The project critiques the role of AI in media representation, challenging viewers to question the authenticity and impact of digitally mediated visuals
HyperChips is a digital art project by Albertine Meunier, created using the DALL·E AI model. The series is based on the playful prompt "Albertine Meunier is eating sausages and chips" and consists of images generated using AI, which she curated and refined to create a set of 303 unique NFTs. Each artwork features a surreal depiction of women engaging with the mundane activity described, highlighting the imperfections and humorous elements that often emerge from AI-generated visuals.
The project has been exhibited at major art venues, including the Grand Palais Éphémère in Paris and other digital art shows, and is a part of Meunier’s exploration of AI-generated imagery as a means to critique and reinterpret traditional representations. More information
Infinite Images by Holly Herndon and Mat Dryhurst is a groundbreaking AI art series created using DALL·E 1. The project was commissioned by OpenAI in 2021 and explores the possibilities of creating large-scale, internally coherent images through a unique patchwork technique. Each artwork begins with a single image that is then "extended" in various directions using the model’s outpainting capabilities, resulting in compositions that can expand infinitely.
The project includes 682 unique artworks and plays with visual themes like water reflections, horizon lines, and surreal landscapes, alluding to narrative art forms like graphic novels and tapestries. The artworks were created by manually guiding the AI’s generation process to ensure coherence and style consistency across expansive canvases, making them some of the largest early AI-generated compositions of their kind.
This series marks a significant point in the evolution of AI-assisted visual storytelling, highlighting the capacity of AI tools to support continuous artistic exploration and expansion.
In"Journey Through the Layers of the Mind" by Memo Akten is an experimental video artwork created on 3rd July 2015 using the DeepDream algorithm, which was one of the first AI image synthesis techniques popularized by Google. The project is considered one of the earliest examples of using neural networks for artistic purposes, visualizing the inner workings of artificial neural networks.
In this piece, Memo Akten used his own video footage as the source material and applied the DeepDream algorithm to transform it into a surreal, psychedelic visual experience filled with fractal patterns, morphing shapes, and hallucinatory images. The work explores how AI “sees” and interprets visual data, blending artificial and biological neural networks in a complex interplay of perception. The resulting visuals often resemble creatures and forms that don’t exist in reality, revealing the biases and interpretative nature of AI.
Sasha Stiles is a first-generation Kalmyk-American poet, artist, and AI researcher known for her innovative exploration of generative literature and blockchain poetics. Her work focuses on the intersection of text, technology, and human experience, blending traditional poetry with cutting-edge AI and digital art. She co-founded theVERSEverse, an experimental crypto literary collective, and has collaborated with AI models like OpenAI’s GPT-3, which she trained on her own poetry to create her "AI alter ego" named Technelegy.
Her major projects include the acclaimed poetry collection Technelegy and series such as Cursive Binary, which merges human handwriting and binary code to reflect on language as a form of encoding human experience. Stiles has exhibited at venues like the Kunsthalle Zurich, Christie’s, Art Basel Miami Beach, and has been recognized as a pioneering figure in digital literature and art. More information
Mario Klingemann is a pioneering German artist who has been at the forefront of AI art since 2015, specializing in the use of neural networks, GANs (Generative Adversarial Networks), and custom machine learning models to create visual art that explores themes of creativity, perception, and the intersection of human and machine. His works have been featured at major art institutions like MoMA, the Centre Pompidou, and the Hermitage Museum.
"Verses/Distributed Consciousness" by Memo Akten is a multi-layered art project that explores themes of consciousness, artificial intelligence, and interspecies connection. Created in 2021, it combines AI technologies such as Generative Adversarial Networks (GANs), CLIP, VQGAN, and GPT-3 to generate visual and textual content inspired by the concept of distributed cognition.
The installation uses eight-channel video, LED strips, and custom software to produce immersive visuals that metaphorically represent distributed intelligence, using cephalopods (octopuses) as a key motif. The project is based on the idea that octopuses, with their unique neural structures spread across their arms, challenge our traditional understanding of centralized intelligence. This metaphor is extended to the rise of AI and distributed computation, reflecting how consciousness can emerge through complex systems in nature and technology alike.
The AI-generated verses created using GPT-3 are cryptographically hidden within the visuals and act as a commentary on topics such as free will, ecology, life, and death. These verses are not just artistic elements but serve as a bridge between human thought and machine learning, making the project a meditation on how technology shapes our perception and narrative of consciousness.
U-Net
Hannu Töyrylä
Unstable Views July 2024
Hannu Töyrylä is a Finnish visual artist who combines AI and digital techniques to explore complex themes of perception and change. His work, "Unstable View," uses a variety of AI tools to create dynamic cityscape images that evolve and transform gradually, challenging the viewer's perception of the scene. He utilizes U-Net models for damaging and restoring images, neural style transfer, and diffusion techniques to achieve a semi-abstract aesthetic.
Linda Dounia Rebez is a Senegalese multidisciplinary artist whose work integrates AI technology, specifically GANs (Generative Adversarial Networks), to explore themes like identity, memory, and the implications of technology on marginalized communities. Her projects often address how AI models reflect biases inherent in the datasets they are trained on and how these biases shape cultural narratives.
Rebez uses AI as a tool for "speculative archiving," creating projects that critique how generative AI models, often trained on biased datasets, can erase or distort the reality of underrepresented cultures. Her practice aims to reclaim agency over the digital memory of communities from the Global South, challenging the often Western-centric biases in AI models. Her work highlights how the exclusion of data from diverse cultural backgrounds leads to incomplete or distorted representations in AI-generated art, making her contributions both artistic and socially critical.
The project was created by Jacqui Kenny, whose agoraphobia limits her ability to physically explore the world. Turning to digital mapping as a way to navigate new places, she discovered "Argelton" and was drawn to its enigmatic nature—a place that exists and yet does not. Inspired by her own experiences of restricted movement and the fragmented perception of space, she used this phantom town as a metaphor for exploring themes of inaccessibility and the blurred lines between what is real and what is imagined. Through Holmes of Argelton Lane, Kenny transforms the constraints of her condition into a creative lens that redefines the way we understand the digital and the physical realms.
Seeds is a digital art series by Yuma Kishi that explores the intersection of nature and technology, visualizing how organic forms can be reinterpreted and transformed through digital processes. Each work in the series is generated using AI-driven algorithms that mimic the growth patterns found in biological structures, producing intricate visuals that resemble seeds, cells, or microorganisms. These digital "seeds" evoke different stages of life—from germination to expansion—capturing a delicate balance between chaos and order. Kishi’s approach highlights his interest in using AI not merely as a tool, but as a form of "alien intelligence" that offers new ways of perceiving and representing the natural world. By translating organic processes into abstract digital forms, he encourages viewers to reflect on the coexistence and potential convergence of nature and technology in the modern era
Helena Sarin is an AI artist and software engineer known for blending generative art techniques with traditional art forms. Her background in technology, including her work at Bell Labs and her experience in computer vision, provides a strong foundation for her artistic explorations. She began her career in AI art using generative adversarial networks (GANs) and developed her own approach, which she describes as “Folk AI.” Sarin often trains AI models on her own drawings, watercolors, and photographs, resulting in a distinct visual language that merges digital aesthetics with handcrafted qualities.
Sofia Crespo's [[[vessel_within_vessel_0099]]] reimagines the concept of a photomosaic through a generative lens, exploring the duality of images and their perception. The artwork challenges traditional notions of resolution and coherence by examining how every pixel—typically seen as a simple square of color—can itself contain a coherent, recognizable entity when magnified. This recursive examination of an image blurs the boundaries between micro and macro perspectives, offering a richly layered visual experience.
By employing a self-trained StyleGAN, Crespo achieves intricate compositions where each pixel or tile within the mosaic not only contributes to the broader image but also holds its own as a standalone micro-entity. This method allows for a seamless integration of local and global coherence, ensuring that the generated micro-images align meaningfully with the overarching composition. The flexibility of the StyleGAN architecture supports this balance, enabling a dynamic interplay between abstraction and detail.
Simultaneously, the photomosaic framework serves as a conceptual and aesthetic tool. It abstracts the original image into a new, stylized form, where each small unit acts as a "vessel" for intricate details and independent meaning. This innovative use of generative AI allows Crespo to merge traditional artistic forms with computational innovation, creating a multi-scale perspective that invites viewers to explore the relationships between the parts and the whole.
Sofia Crespo is a prominent artist in the field of AI-driven art, known for her bio-inspired creations that blend natural and synthetic elements. Using custom-trained AI models like StyleGAN, she transforms datasets of biological patterns and textures into intricate, surreal representations of imagined ecosystems. While Crespo focuses on adapting and training existing AI models rather than developing them, her work redefines the boundaries of creativity by simulating and expanding upon biological processes. Through projects like Artificial Natural History and vessel_within_vessel_0099, Crespo invites audiences to reconsider the relationship between technology and the natural world, emphasizing the evolving role of AI in artistic exploration.
Golan Levin is an American artist, engineer, and educator known for his pioneering work at the intersection of art and technology. His practice focuses on interactive new media, combining visual art, computational design, and human-machine interaction to explore expressive, nonverbal communication through digital mediums. He has been a significant figure in software art since the 1990s and is recognized for his unique approach that blends whimsy, provocation, and cultural critique.
Levin is currently a professor at Carnegie Mellon University, where he teaches courses on interactive art and generative design. His career spans work in high-tech research environments like the MIT Media Lab and the Ars Electronica Futurelab. He has exhibited widely at renowned institutions such as MoMA, the Whitney Biennial, and Ars Electronica, and his work is part of the permanent collections at several major museums.
Agoria (Sébastien Devaud) is a French multidisciplinary artist, music producer, and pioneer in “biological generative art,” which integrates elements of technology, nature, and artificial intelligence. His work merges visual art, science, and music, exploring themes such as human gesture, biological systems, and generative algorithms. Agoria collaborates with scientists and philosophers to create pieces that reveal the beauty of the unseen patterns in both natural and digital environments.
His art often bridges the gap between physical and digital realms, as seen in projects like “Le Code d’Orsay” at the Musée d'Orsay, where he combined digital art with scientific explorations to reinterpret classical works through a generative lens. Agoria’s projects have been exhibited at prestigious venues like the Tate Modern and Miami Art Basel.
Aurèce Vettier is a French art project founded in 2019 by Paul Mouginot. It blends art, AI, and technology to push the boundaries of creative processes by merging formal research, generative algorithms, and craft techniques. The name "Aurèce Vettier," created using an algorithm, symbolizes a collaborative and open approach to art-making that fluctuates between the tangible and digital realms.
The project's works range from digital art and poetry to bronze sculptures, emphasizing the interplay between traditional artistic methods and algorithmic processes. Some of Aurèce Vettier’s notable series include the “Potential Herbariums,” which represent AI-imagined botanical forms, and explorations in virtual and physical mediums that challenge the distinction between human and machine-generated aesthetics
Entangled Others is an experimental artist collective formed by Sofia Crespo and Feileacan Kirkbride McCormick. Their practice focuses on exploring the complex relationships between ecology, technology, and art, with a particular emphasis on the interconnectedness of all life forms—both human and non-human. The duo's work investigates the concept of entanglement, highlighting how no single entity can exist in complete isolation and how every action reverberates through a web of interconnected beings.
Their art, which blends generative and digital methods, aims to create new forms that give presence and life to the "more-than-human" world in digital space. By using AI and biology-inspired technologies, they encourage viewers to rethink the boundaries between natural and artificial life. Their projects often address themes such as biodiversity, technology’s impact on ecological systems, and the redefinition of creative processes using generative techniques.
Chimerical Stories is a groundbreaking exploration of StyleGAN2 (Generative Adversarial Networks), where the latent spaces of two neural network models are blended to create hybrid, surreal specimens. The process employs a layer blending technique, combining the higher layers of one model—responsible for global shapes and structural forms—with the lower layers of a second model, which generate fine-grained details like textures and intricate patterns. This "crossbreeding" of latent spaces parallels genetic recombination, merging traits from datasets representing distinct species and habitats.
The resulting hybrid models were navigated and refined using custom code designed to extract genetic-like traits such as form, color, coherence, and environmental characteristics. These traits were further manipulated to explore their interactions, ultimately producing unique images and videos. Each creation in Chimerical Stories reflects a unique crystallization of blended traits, embodying a rich, aquatic, and dreamlike aesthetic.
The series relies on StyleGAN2, a state-of-the-art GAN architecture that offers precise control over abstraction levels within the latent space. The advanced Style Mixing capabilities of StyleGAN2 allow for seamless blending of structural and stylistic attributes, while its reduction of visual artifacts ensures the creation of highly detailed and coherent outputs. Latent space exploration was central to the creative process, adding layers of richness and complexity to the generated specimens.
Xander Steenbrugge is a Belgian generative artist, AI researcher, and founder of the digital media platform WZRD.ai. His work focuses on exploring the creative potential of AI through generative models and machine learning. He integrates these technologies to create complex, visually immersive art pieces that blend sound and imagery, aiming to push the boundaries of human-machine collaboration in artistic creation.
Steenbrugge is known for projects like "Neural Synesthesia," where he uses AI models to create art that is synchronized with music, producing dynamic visuals based on audio feature extraction. He is also a prominent public speaker and educator, sharing insights on AI’s impact on creativity and society through his YouTube channel, “Arxiv Insights”.
In addition to his digital art, Steenbrugge is involved in the startup Eden, which empowers artists to create personalized AI models, enhancing their creative processes and expanding the possibilities of generative art.
Memo Akten is a Turkish artist, researcher, and computational artist known for his interdisciplinary approach that blends art, science, and spirituality. He explores themes such as human perception, consciousness, and the intersection between natural and artificial systems. His work often involves creating data-driven visualizations and behavioral abstractions that investigate complex processes in nature and human behavior.
Akten holds a PhD in Artificial Intelligence with a focus on expressive human-machine interaction from Goldsmiths, University of London. His projects use a variety of digital and generative techniques, often involving AI to create immersive installations, interactive performances, and visual compositions. He has exhibited at major venues like the Barbican, the Grand Palais, and the Moscow Museum of Modern Art and has received several awards, including the prestigious Golden Nica at Prix Ars Electronica.
Akten’s artistic practice is deeply inspired by natural processes and fundamental scientific questions, aiming to create unfamiliar forms that evoke a deeper understanding of our world. He currently serves as an Assistant Professor of Computational Arts at the University of California, San Diego
Anne Spalter is an American digital artist, academic pioneer, and author, widely recognized for founding the first digital fine arts programs at Brown University and the Rhode Island School of Design (RISD) in the 1990s. Her work blends traditional art techniques with computational processes, using custom software and artificial intelligence to explore modern landscapes and futuristic narratives. Spalter is also known for her role as a collector and advocate for digital art, establishing one of the largest collections of early computer art with her husband, Michael Spalter.
One of her prominent AI art projects is the “AI Spaceships” series. This collection consists of 501 unique, AI-generated spaceship artworks, each depicting surreal, sci-fi-inspired scenes. The narrative envisions a future where climate change has rendered Earth uninhabitable, prompting humanity to build spaceships in a desperate attempt to escape. These ships embark on complex journeys, some encountering failures, while others warp through time or confront new, bizarre lifeforms. Spalter created the series using text-to-image AI models, adding her own artistic touches through digital post-processing to create visually captivating and thought-provoking pieces.
Dreamwalking
Cyperpunk vision
The final hour
Art explosion
Mike Tyka is a multi-disciplinary artist and scientist known for his innovative use of deep neural networks to create digital art. With a background in biochemistry and a PhD in biophysics, Tyka initially worked in scientific research, studying the structure and dynamics of protein molecules. His journey into the art world began in 2009 with a large-scale, functional art installation called "Groovik's Cube," a multi-player, interactive Rubik’s Cube. Later, Tyka co-founded the Artists and Machine Intelligence program at Google, which marked the beginning of his focus on blending AI and art.
Total isolation
I am the universe
Emergent Finale 00181
ContentBeware
"Almost Human" by Irina Angles and Dr. Formalyst is a collaborative digital artwork that delves into the blurred lines between human emotion and artificial cognition. The piece portrays a humanoid figure, partially deconstructed, symbolizing the fragmented nature of identity in the digital age. The central motif is a hybrid form: part human, part machine, yet distinctly otherworldly.
Angles and Dr. Formalyst use a combination of generative algorithms, 3D modeling, and neural networks to create a visual narrative that challenges our perception of what it means to be human. The work's aesthetic is characterized by soft, organic textures contrasting with sharp, mechanical elements, highlighting the duality between natural and artificial.
Thematically, “Almost Human” questions the growing role of AI in society and its impact on self-perception. It evokes a sense of uncanny familiarity, making viewers reflect on their own relationship with technology. The gaze of the figure—empty yet compelling—suggests a yearning for something more: a search for identity in a realm where boundaries between the self and the machine are increasingly indeterminate.
Encoder-Decoder-Network
"Learning to See" by Memo Akten is a 2017 digital art series that explores how machines perceive and reconstruct visual information. Using Convolutional Neural Networks (CNNs), the piece simulates a machine’s interpretation of visual data, having been trained on specific patterns and textures. When given new inputs, the network attempts to reconstruct these images according to its learned biases, resulting in surreal, fragmented visuals that reflect its internal understanding.
The work serves as a reflection on the subjective nature of perception, illustrating that both human and machine vision are influenced by prior experiences and training. Through this distorted, dreamlike lens, Akten emphasizes the limits of machine intelligence, challenging our assumptions about artificial perception and its impact on how we see and interpret reality.
0RAL B1NARY
CURSIVE BINARY
EVERY POEM STARTS WITH A SEED
"Measured Confrontation at Formica Feast" (created with Stable Diffusion XL) Where gazes unravel the threads of discourse, hushed voices dissecting the fragments of lore. Shapes and patterns but shadows of source, in measured confrontation, the mind's silent war. Stilled by the weight of collected abstraction, curators of thought in their intricate dance. Blue-coated scholars in tacit interaction, seeking the keys to chance's expanse. To dine on ideas, a feast of inception, savoring nuances never quite caught. Cerebral repast, the banquet's deception, with courses of concepts eternally sought. In chambers of intellect, echoes of reason, unspoken questions in every glance. Dissecting the chaos with surgical precision, a measured confrontation, the scholars' stance. Agents of clarity in a cavern of query, where knowledge is served on the platters of time. Nibbling on scraps of the theoretical quarry, with every morsel, a step in the climb. No spoon or fork for the feast they prepare, just the cutlery of consciousness, sharp and defined. In each piece they parley, a layered affair, where the genesis of form is inherently twined.
Mare Nostrum
人工アイ像
Botto
Botto is an autonomous AI artist co-created by Mario Klingemann that continuously generates thousands of images based on text prompts using a blend of AI models, including VQGAN + CLIP, Stable Diffusion, and custom text-to-image frameworks. Botto’s creative output is directed by community feedback through the BottoDAO, where members vote weekly on their favorite images. Central to this system is the Botto Token, a governance token that enables holders to participate in these votes and shape Botto’s artistic development. By engaging a decentralized network of contributors, Botto redefines authorship in digital art, with its works showcased internationally as a testament to the collaboration between human feedback and machine intelligence.
The Rapture
we are the trees on a new planet, the day after the last human left for the stars
aurèce vettier (Paul Mouginot)
we are the trees
Ryan Murdock (Advadnoun)
A Bearded Man Befrieding A Forest
le travail des rêves
aurèce vettier (Paul Mouginot)
Vladimir Alexeev (Merzmensch)
Vladimir Alexeev, also known as Merzmensch, explores the intersection of AI and human memory in his artwork MERZmory Diffused. In this project, he uses neural networks to reconstruct visual memories from his personal photographs. The work highlights the creative potential of machine "failures," where the AI's inability to perfectly replicate memories generates new, abstract aesthetic forms. By embracing these imperfections, Alexeev expands the boundaries of human-machine collaboration, crafting a unique dialogue between memory, technology, and art.
Ryan Murdock, also known by his pseudonym Advadnoun, is a well-known figure in the AI art community for his contributions to AI-driven generative art. He is credited with developing some of the earliest VQGAN + CLIP notebooks, which have had a significant influence on the text-to-image generation space. His work helped to pioneer the method of combining VQGAN (Vector Quantized GAN) for image generation with CLIP (Contrastive Language-Image Pretraining) to guide these images based on textual prompts.
In the project MERZmory by Vladimir Alexeev, known as Merzmensch, StyleGAN plays a critical role in exploring the boundaries of human memory and AI's ability to reconstruct or interpret it. MERZmory uses neural networks to generate visual "memories" based on datasets curated by Merzmensch, such as his personal photographs. StyleGAN is employed in this context to produce images that are both familiar and surreal, capturing the essence of memory's fluid, sometimes distorted nature.
In MERZmory, the GAN framework allows the AI to generate visual outputs that mimic how human memory can be reconstructed or even altered over time. StyleGAN's capability of controlling different levels of image synthesis is particularly useful here, as it enables fine manipulation of the visual elements—analogous to how memories evolve or degrade in our minds. The "hallucinations" generated by the AI are viewed as creative failures, producing novel, imaginative forms that go beyond the original input, which aligns with the Dadaist philosophy that Merzmensch often explores in his work.
Vladimir Alexeev (Merzmensch)
8 Years
#MERZmory Diffused
Spectacles of time
me & me
Clouds
Noper (bagdelete)
don´t buy
Day #35: Collecto_ERros
Day #60: Forever Motions
Postino.ai by Chikai
Botto
Measured Confrontation at Formica Feast
The Genesis series, launched in 2021, is Botto’s first collection, produced entirely with VQGAN + CLIP. This foundational series captures the AI artist's early exploration of abstract and surreal themes, with compositions characterized by bold colors and unexpected visual forms. Through early community voting, Genesis set the tone for Botto’s future stylistic direction, helping to refine the AI’s understanding of aesthetic values and marking the beginning of a collaborative creative process between machine and community.
Exorbitant Stage is a key artwork in Botto’s Genesis series, created during the AI artist’s first year of production. Minted in October 2022, this piece represents the culmination of Botto’s early explorations in digital art using open-source text-to-image models. Thematically, Exorbitant Stage reflects on the concept of theatricality and its costs, with elements like figurative "actors," symbolic bouquets for an audience, and detailed "costumes" that evoke the idea of a staged production. This work invites viewers to consider the spectacle and artifice inherent in both performance and digital art creation.
This artwork holds a special place within Botto's catalog as one of the last mints of the Genesis period before the transition to more advanced models like Stable Diffusion. Exorbitant Stage was minted just before Botto’s "Fragmentation Period," marking a transformative phase as Botto began to utilize new technologies and models, reshaping the possibilities for digital and AI art.
Botto’s eighth period, themed around “Morphogenesis,” delves into the human form, oscillating between high realism and abstraction. Traditionally, morphogenesis describes the biological process through which organisms shape and form, like the development of a human embryo. In Botto's interpretation, morphogenesis explores how form and structure emerge from chaos, symbolizing creation and transformation. This theme reflects Botto's creative journey, merging AI-driven guidance with human collaboration. The period emphasizes the distinct line between AI and organic forms, drawing from and examining the physical essence of human embodiment.
Period #8: Morphogenesis
Period #1: Genesis
Period #9: Synthetic Histories
Botto’s ninth period, Synthetic Histories, explores history as a construct shaped by human choices and biases. Like AI-generated content, history is often a product of selective memory, reliant on archivists and historians whose interpretations may reflect subjective truths. Botto, acting as a synthesizer of history, reimagines past events through the lens of artificial intelligence, questioning how reliable and stable our understanding of history truly is.
This period prompts reflection on AI's emerging role as a curator and creator of historical narratives. As it synthesizes vast amounts of data, AI challenges the notion of a singular, authoritative past, instead presenting history as fluid and open to reinterpretation. Synthetic Histories invites us to consider: How does AI reshape our perception of history? Could its reinterpretations challenge the authenticity of human memory, and how will our collective beliefs evolve as AI begins to shape how we remember and define the past?
Past Forward
Alkan Avcıoğlu
All Watched Over By Machines Of Loving Grace
Botto's seventh period, titled "Temporal Echoes," delves into the multifaceted nature of time, exploring its cyclical and linear perceptions. The term "temporal" originates from the Latin temporalis, meaning "of time," while "echo" stems from the Greek ēkhē, or "sound." In this period, Botto investigates how past narratives and future possibilities intertwine, creating artworks that reflect the continuous dialogue between history and potential futures. This exploration is a testament to Botto's evolving artistic journey, where AI-driven creativity merges with human collaboration to produce pieces that resonate with both nostalgia and foresight.
Period #7: Temporal Echoes
"Neon Whispers in Parallel" (created with Stable Diffusion XL) The city never truly sleeps, not in the traditional sense. Its consciousness merely shifts, wandering through cycles of vivid clarity and hazy dreams. The streets of the urban labyrinth, veins filled with the lifeblood of relentless activity, begin to exhale the day's frenzy as dusk falls. It's amidst this transformation that the neon begins to whisper.
Alex strolled down the boulevard, as the first hints of the synthetic auroras flickered to life. The streetlights, elders of the cityscape, bore witness to his nightly pilgrimage. Their amber eyes reflected in the puddles left by the afternoon rain, pools of captured sky on the stained concrete canvas.
Encased by glowing signs and storefronts that clung to the last patrons of the day, Alex observed the silent conversations of passersby—silent to the ear but loud in gesture and gaze. He found solitaire comfort knowing each silhouette encased untold stories. Tonight, the whispers grew louder, a symphony composed of the hums of turning neon signs, the rhythmic clatter of a distant train, and the soft murmur of a city settling into its nocturnal embrace.
Neon Whispers in Parallel
Exorbitant Stage
Sultry Nights, The Backyard Chronicles, 2020
David Young
David Young is a contemporary American artist who explores how artificial intelligence can be used to reinterpret and extend our understanding of the natural world. His work is rooted in the idea that technology, while often seen as separate from nature, can offer new perspectives on the organic patterns and beauty around us. Through his art, Young examines how machines learn to “see” and “reimagine” nature, leading to unique, often surreal reinterpretations of familiar natural forms.
Learning Nature is one of Young's pivotal series, where he trained AI models to observe, learn from, and recreate images of natural forms. This series began with Young photographing flowers and other organic textures at his farm in Bovina, New York, over the summer of 2018. These photographs served as the data source for the machine learning models, which he trained to generate new images inspired by these natural scenes.
The results are images that feel both familiar and alien. The AI interprets and recreates natural structures, but it also introduces its own "creative errors" — distortions, abstract patterns, and unfamiliar shapes that emerge as the AI's interpretation of the natural data. This process raises fascinating questions: How does AI perceive the world differently from humans? Where does nature end, and where does the machine’s creativity begin?
Symbiosis
Harold Cohen
AARON is a landmark computer program developed by Harold Cohen in the early 1970s, designed to autonomously create artworks. Initially producing black-and-white line drawings, AARON evolved over decades to incorporate the use of color, intricate compositions, and elements of abstraction. The program was designed to simulate artistic decision-making, relying on rules and algorithms written by Cohen, rather than randomness. AARON was not just a tool but a creative system that demonstrated the potential of artificial intelligence in art. It remains a milestone in the history of digital and generative art, blurring the lines between human and machine creativity.
Harold Cohen (1928–2016) was a British-born artist and one of the pioneers of computer-generated art. Initially a painter, Cohen began exploring the potential of computers for creative processes in the late 1960s. Fascinated by the intersection of art and technology, he devoted much of his career to developing systems that could autonomously produce art. Cohen’s work challenged traditional notions of authorship and creativity, emphasizing the collaborative relationship between the artist and the machine. His groundbreaking achievements earned him recognition as one of the founding figures in the field of computational creativity.
AARON
This work forms part of the 2002-2004 Digital Prints Era which marked a pivotal moment in Harold Cohen’s artistic journey. Many were shown at the ‘Untouched by Hands’ exhibition at the Earl & Birdie Taylor Library in San Diego, California. They were originally printed as larger-scale works on the artist’s wide-format Roland printer. The works highlight the artist’s evolving approach to representation, as AARON reimagined natural forms—particularly potted plants—and experimented with their arrangement in space, from tables and tiled floors to their eventual liberation from pots altogether. The collection represents a crucial departure from earlier representational forms, with Cohen allowing AARON to autonomously manipulate these elements, creating a seamless blend of human input and machine-generated imagery.
0309-03s (C3 AW50)
For the first time, the Harold Cohen Trust is pleased to make available a selection of his works posthumously minted as ERC-721 tokens on the blockchain.
The AARON program generated a vector file in a custom format designed by the artist called the ‘AA’ format. The program generated both a vector file for the black line drawing and also a color file. These two different files were merged together in the ‘AA” format which further included additional information about how to make the artwork. These ‘AA’ files were referred to as the AARON Artwork descriptor files and were used to create the JPEGS.
ClownVamp
The Junk Machine
This striking artwork merges nature, technology, and artistic innovation. Captured in Central Park, New York, and later transformed using advanced AI techniques, Azalea Walk Dreamscape is a pioneering example of computational photography reimagined through artificial intelligence.
Originally commissioned as a large-scale light box for the 2016 “9e2” art fair in Seattle, which celebrated the intersection of art and technology, the piece toured notable venues, including Seattle City Hall and Google’s NYC flagship office. It also served as the wraparound cover for Ian Goodfellow's influential textbook Deep Learning, reflecting Ambrosi's groundbreaking contributions to AI-driven art.
Azalea Walk Dreamscape stands as a testament to the evolving relationship between human creativity and machine intelligence, inviting viewers to explore the potential of these powerful tools in reshaping artistic expression.
Daniel Ambrosi
Azalea Walk Dreamscape
Daniel Ambrosi is a pioneering artist whose work explores the intersection of art, technology, and nature. Known for his innovative use of computational photography and AI, Ambrosi creates dreamlike landscapes that reimagine natural scenes through advanced digital techniques. His art has been featured in prominent exhibitions, corporate spaces, and as the cover of Ian Goodfellow's influential textbook Deep Learning. Ambrosi's work challenges traditional boundaries, offering a glimpse into the future of creativity powered by technology.
Holly Herndon and Mat Dryhurst are Berlin-based artists and innovators known for their groundbreaking work at the intersection of music, artificial intelligence, and decentralized technologies. Together, they explore the creative and ethical potential of AI, including projects like Spawn and Holly+, which use machine learning to reimagine vocal performances and music composition.
In addition to their artistic endeavors, they are deeply involved in the development of decentralized systems, advocating for artists' rights and exploring new models for creative ownership through DAOs (Decentralized Autonomous Organizations). Their influence extends to prominent exhibitions and discussions, making them central figures in the evolving conversation about art, technology, and collaboration in the digital age.
Milford Sound Infinite Dream
Created in 2020, the Infinite Dreams series merges the precision of photography with the creativity of artificial intelligence, specifically the DeepDream algorithm. Large-scale landscape images are reimagined into a visual language reminiscent of Cubism: surfaces and shapes fragment, overlap, and form a new dimension of reality. The AI transforms familiar scenes into surreal dreamscapes, bringing details to life and inviting viewers into a world where reality and digital imagination converge.
Alexander Mordvintsev is a computer scientist and experimental artist best known as the creator of Google’s DeepDream algorithm. His work lies at the intersection of artificial intelligence, computer vision, and creative expression. By exploring the capabilities of neural networks to generate surreal, dream-like images, Mordvintsev has helped redefine the role of AI in art and visual storytelling.
Through his practice, Mordvintsev investigates the relationship between human perception and machine creativity. His art often emphasizes the latent patterns and structures hidden within data, revealing how machines interpret and transform visual information. By blending computational techniques with aesthetic exploration, he inspires viewers to reconsider the boundaries between human and machine-generated creativity. His work has influenced both the artistic and scientific communities, sparking new conversations around the potential of AI as a collaborative creative tool.
Elman Mansimov is a computer scientist and innovator in the field of generative AI, widely recognized for his contributions to the development of text-to-image generation models. His pioneering work with neural networks, including the alignDRAW model, has laid the foundation for modern AI-driven image synthesis techniques.
Mansimov's practice explores the intersection of machine learning and creative expression, focusing on how natural language can be transformed into visual representations. By integrating neural attention mechanisms and advanced generative algorithms, he has opened new pathways for collaboration between humans and machines in artistic creation. His work challenges traditional notions of authorship and creativity, inviting audiences to consider the evolving role of AI in shaping the future of art and technology.
From The Redwood Forest
In Evolved Hallucinations, Trevor Paglen challenges the reductive literalism of machine vision systems. While such systems often interpret images in rigid terms, Paglen explores how "seeing" is shaped by history, culture, subjectivity, and biology.
For this project, Paglen trained AI models on datasets from allegorical art, philosophy, folklore, and metaphor to simulate alternative perspectives—such as viewing the world through the eyes of a post-human future or a Cassandra-like figure fated to foresee but unable to change events.
These works offer a new lens on perception, highlighting that seeing is more than recognition—it is a deeply cultural, emotional, and political act. Paglen’s project invites reflection on the boundaries and possibilities of machine vision.
Trevor Paglen (*1974) is an American artist, researcher, and geographer known for his investigations into surveillance, artificial intelligence, and the invisible infrastructures that shape our world. His interdisciplinary work spans photography, sculpture, and installation, often uncovering hidden systems of power and control.
Paglen’s practice combines rigorous research with artistic experimentation, using tools from fields like computer science and cartography to explore themes of visibility and perception. From documenting classified military sites to exposing the biases of AI systems, his work challenges audiences to reconsider the technological and political forces that shape their lives.
Paglen has exhibited internationally at renowned institutions, including the Smithsonian American Art Museum, the Tate Modern, and the San Francisco Museum of Modern Art. In 2018, he collaborated with scientists to launch Orbital Reflector, a satellite artwork, into space. Through projects like Evolved Hallucinations, Paglen continues to push the boundaries of how we understand and experience the world.
Bård Ionson is a trailblazer in generative and crypto art, merging AI, blockchain, and traditional media to create works that explore the boundaries between physical, digital, and spiritual realms. Since 2018, he has used tools like GANs and smart contracts to push artistic innovation. His globally exhibited works challenge the relationship between humans and machines, offering fresh perspectives on the role of technology in art.
Alkan Avcıoğlu is a multidisciplinary artist whose work bridges cinema, digital art, and AI post-photography. His latest series, All Watched Over by Machines of Loving Grace, critiques the dehumanizing effects of technology, portraying humanity as overwhelmed by screens and consumerism. Drawing on his background as a film critic, DJ, and academic, Avcıoğlu blends narrative depth with visual excess, reflecting on themes of alienation, hyperreality, and post-capitalist culture. His art challenges traditional authorship, celebrating randomness and collective unconsciousness in the digital age. From urban density to AI-crafted images, his work synthesizes diverse influences into a compelling critique of modern life.
Neural networks are the foundation of many AI-generated artworks, enabling systems to analyze data, recognize patterns, and create independently. Recurrent Neural Networks (RNNs) are particularly important for processing sequential data such as text, music, or image sequences. Their ability to "remember" previous inputs makes them ideal for narrative art projects, where context and continuity are crucial—such as creating cohesive visual stories or animating images.
In AI art, RNNs are often combined with other network types. Convolutional Neural Networks (CNNs) excel at analyzing visual patterns, identifying shapes, colors, and textures. Generative Adversarial Networks (GANs) produce entirely new artworks by pitting two networks against each other: one generates images, while the other evaluates them, resulting in increasingly realistic and creative outputs. Building on this, VQGAN + CLIP combines image generation with semantic understanding, creating works that visually explore intricate meanings and concepts.
Diffusion models, on the other hand, generate images by transforming initial noise into detailed and multi-layered visuals, offering unparalleled precision and artistic depth.
Recurrent Neural Networks (RNNs) are designed to process sequential data by retaining information from previous steps, making them ideal for tasks requiring temporal or contextual understanding. A key innovation in their application to AI art is their ability to iteratively generate and refine content, creating coherent and meaningful outputs.
One notable example is AlignDRAW, which integrates RNNs with an attention mechanism to generate images conditioned on text descriptions. Built on the DRAW architecture, AlignDRAW uses a recurrent process to "paint" images in multiple steps, much like an artist refining a canvas layer by layer. By incorporating LSTM (Long Short-Term Memory) units, the model overcomes the limitations of traditional RNNs, effectively managing long-term dependencies across iterations. The addition of an attention mechanism allows AlignDRAW to focus on specific parts of a text description when generating corresponding image elements, ensuring alignment between the textual input and visual output.
AlignDRAW showcases how RNNs can power iterative and context-aware creativity in AI art. While modern models like diffusion networks have advanced these capabilities, RNN-based architectures remain a foundational step in understanding and leveraging sequential generation processes in art.
Convolutional Neural Networks (CNNs) are specialized neural networks designed for visual data processing. They analyze images hierarchically, identifying simple patterns like edges in early layers and complex structures like objects in deeper layers. This makes them essential for tasks such as style transfer - which often uses an encoder-decoder architecture to combine the style of one image with the content of another—and for generating surreal, dreamlike visuals in projects like DeepDream, where CNNs amplify patterns to create intricate, abstract artworks.
CNNs are also a core component in models like Generative Adversarial Networks (GANs), where they are used to both generate and evaluate images. Their ability to deeply understand and manipulate visual content has made CNNs a cornerstone in AI art, enabling everything from realistic image synthesis to highly creative, stylized outputs.
DeepDream, created by Alexander Mordvintsev during his time at Google, uses Convolutional Neural Networks (CNNs) to visualize how AI "sees." By amplifying features the network strongly responds to, it generates surreal, psychedelic images filled with exaggerated patterns like eyes, spirals, or dreamlike landscapes.
Initially developed as a tool to understand the inner workings of CNNs, DeepDream has since become a cornerstone of AI-generated art, blending scientific exploration with creative expression.
Encoder-decoder networks are neural architectures designed to transform input data into a compressed representation and then reconstruct it into a desired output format. The encoder processes the input, such as an image or text, and extracts key features, reducing the data to a compact, latent representation. The decoder then uses this latent representation to reconstruct the output, often transforming the input into a new form, such as a stylized image or a translated text.
In AI art, encoder-decoder networks are commonly used for tasks like style transfer, where the encoder captures the content of one image, and the decoder integrates the stylistic features of another to create a visually compelling hybrid. They are also applied in image denoising, super-resolution, and text-to-image generation, enabling creative transformations while preserving essential details. By compressing and reconstructing data efficiently, encoder-decoder networks form the foundation for many artistic and functional AI applications.
U-Net is a convolutional neural network originally developed for biomedical image segmentation but now widely used in various image processing tasks. Its "U"-shaped structure consists of an encoder path, which compresses the input image by extracting hierarchical features through convolution and pooling layers, and a decoder path, which reconstructs the image using upsampling layers. A defining feature of U-Net is its skip connections, which link the encoder and decoder, allowing fine-grained details from earlier layers to be combined with broader contextual features from deeper layers. This design enables U-Net to produce highly precise outputs, making it ideal for tasks like segmentation, denoising, and image generation, and a valuable tool in both AI-driven art and scientific applications.
Transformer models are deep learning architectures that excel at processing sequences of data simultaneously, thanks to their revolutionary self-attention mechanism, which captures relationships across entire inputs. Initially designed for natural language processing (NLP), transformers have become the foundation of Large Language Models (LLMs) like GPT, scaling to billions of parameters and vast datasets to enable context-aware text generation by understanding dependencies between words.
Beyond NLP, transformers have evolved into multimodal systems, such as DALL·E, extending their capabilities to combine text and images. In these applications, transformers treat text prompts as sequences and image patches as tokens, enabling the generation of corresponding visual outputs. This adaptability has positioned transformers as a cornerstone of both functional NLP and creative AI, seamlessly bridging the gap between language and visuals while driving innovation across industries.
Natural Language Processing (NLP) enables AI to interpret, generate, and manipulate human language, forming the basis for creative text-based applications in AI art. At the forefront of this evolution are Large Language Models (LLMs), which utilize deep learning and Transformer architectures to process and generate human-like text with remarkable fluency.
Prominent LLMs such as GPT (OpenAI), Claude (Anthropic), Grok (xAI), LLaMA (Meta), Mistral, and Gemini (Google DeepMind) have redefined computational creativity. These models generate poetry, narratives, conceptual text-based art, and even interactive literary works. Artists and writers use LLMs to co-create, explore algorithmic authorship, and experiment with AI-driven storytelling.
Beyond simple text generation, LLMs contribute to AI art by enhancing idea generation, providing linguistic interpretation, and enabling interactive dialogues that redefine the relationship between human and machine creativity. Their role continues to expand, shaping new frontiers in literature, digital media, and computational aesthetics.
Multimodal transformers extend the capabilities of traditional transformers to handle multiple types of input, such as text, images, or audio, within a single model. By combining these modalities, they enable seamless interaction between language and visuals, unlocking powerful applications.
Key to their success is the cross-attention mechanism, which aligns features from different modalities, such as matching text descriptions with image components. Models like DALL·E and CLIP use multimodal transformers to generate images from text prompts or analyze images based on textual input. This technology is widely applied in text-to-image generation, visual search, and interactive AI art, offering a more integrated and flexible approach to machine understanding and creation.
DALL-E
"We Are Trees" is an innovative art series by the collective Aurece Vettier, blending nature, technology, and collective memory. The series explores the deep connections between humans and trees, focusing on the ways in which trees act as living archives of environmental and cultural histories. Through the use of AI technologies like GANs and GPT, the works in "We Are Trees" create poetic reflections on growth, decay, and interconnectedness.
The visual elements are likely generated using GANs, which are ideal for creating the organic, tree-like forms and dynamic compositions that define the series. These generative models are trained on ecological imagery and transformed into unique visual representations. Meanwhile, GPT or related transformer models may have been employed to craft the accompanying narratives or conceptual texts that deepen the interpretive experience of the artworks. This combination of AI models bridges the generative visual and textual domains, creating a holistic and immersive exploration of the theme.
By merging organic inspiration with computational processes, "We Are Trees" offers a poignant commentary on the coexistence of the biological and the artificial. Highlighted in galleries and exhibitions, it exemplifies how AI can be used to reimagine the natural world and its cultural significance through art.
DALL·E is a multimodal AI model developed by OpenAI, first introduced in January 2021. It generates images from text descriptions using a transformer architecture, combining text and visual data to interpret natural language prompts and produce highly detailed, imaginative visuals. DALL·E excels at creating diverse outputs, from realistic depictions to abstract, surreal compositions, based entirely on the input prompt.
Applications of DALL·E include art creation, design prototyping, and visual storytelling, making it a groundbreaking tool for both creative and practical uses. By bridging language and imagery, DALL·E demonstrates the transformative potential of AI in expanding the boundaries of artistic expression.
DALL·E 2, introduced by OpenAI in April 2022, is the successor to the original DALL·E model, offering significant advancements in text-to-image generation. It produces more detailed, accurate, and visually coherent images from natural language prompts. Using improved transformer-based architecture and diffusion models, DALL·E 2 can handle complex descriptions, combine multiple concepts, and create high-resolution outputs.
This enhanced capability has expanded applications in creative fields like art, design, and storytelling, making DALL·E 2 a revolutionary tool for bridging text and visuals with unprecedented quality and creativity.
CLIP (Contrastive Language–Image Pretraining), introduced by OpenAI in January 2021, is a multimodal AI model that connects text and images. Trained on a large dataset of text-image pairs, CLIP learns to associate textual descriptions with corresponding visual representations.
Its core functionality lies in understanding both modalities simultaneously, enabling tasks like image classification, zero-shot learning, and generating images based on text prompts when combined with models like VQGAN or diffusion models. CLIP's ability to bridge language and visuals makes it a cornerstone of modern AI art and creative applications.
CLIP (Contrastive Language–Image Pretraining)
Niceaunties
Niceaunties Day #39
Auntieverse
NiceAunties is an AI-driven art project by a Singaporean artist exploring the vibrant, surreal, and multifaceted world of "auntie culture," a term used in Southeast and East Asia to describe older women. Using tools like MidJourney for text-to-image generation, the project reimagines aunties as adventurers in a fantastical "Auntieverse." Here, they dance on beaches, drive tofu-engineered cars, and study at a lunar cooking academy, all while embracing freedom, humor, and individuality.
The artwork also addresses deeper themes, such as loneliness, environmental destruction, and societal expectations of women. Inspired by the artist’s own family, including her grandmother's life in Singapore, NiceAunties transforms stereotypes into playful, empowering narratives, blending social commentary with joyful, AI-generated visuals. The project has gained international recognition, appearing in exhibitions and auctions, while sparking dialogue on identity, aging, and the role of AI in art.
Roope Rainisto is a Finnish AI artist renowned for his innovative use of generative technologies to create visually striking and emotionally resonant artworks. His work often combines elements of AI-generated imagery with photography, resulting in surreal compositions that explore themes of memory, identity, and the passage of time.
Rainisto gained significant attention with his AI-generated series "Life in West America", a collection of nostalgic, cinematic scenes created using tools like Stable Diffusion. His art blurs the lines between reality and imagination, pushing the boundaries of what is possible with AI in creative expression.
Rainisto’s works have been exhibited internationally and have become a significant point of discussion in the intersection of technology and art, highlighting the potential of AI to expand traditional artistic practices.
Alice Gordon is a contemporary artist who uses artificial intelligence to explore themes of identity, self-expression, and the fluid boundaries between human and machine creativity. Her work often incorporates GANs (Generative Adversarial Networks) and diffusion models to produce striking visual compositions that blend the personal with the surreal, challenging conventional notions of portraiture and representation.
Gordon’s pieces frequently examine the fragmentation of identity in the digital age, drawing on her own experiences and cultural heritage to create deeply introspective narratives. By combining traditional photography with AI-generated elements, she constructs hybrid forms that evoke a sense of both familiarity and alienation.
Her work has been showcased in exhibitions across Europe and North America, receiving critical acclaim for its innovative approach to storytelling and its ability to provoke reflection on the intersection of technology and humanity. Alice Gordon’s art pushes the boundaries of what it means to create in an era where machines can collaborate in the act of expression.
The Junk Machine, created by the artist ClownVamp, is an AI-driven art project that explores the intersection of advertising, consumerism, and artificial intelligence. Using the advanced capabilities of the SDXL Turbo diffusion model, the project transforms mundane "junk mail" into surreal, uncanny artworks that challenge traditional perceptions of communication and media saturation.
Each piece blends absurd visuals with fragmented, cryptic text, creating a dreamlike yet unsettling reinterpretation of everyday advertising. These compositions critique the pervasive influence of consumerism, highlighting the ways technology can both amplify and distort modern communication. By turning disposable artifacts of daily life into thought-provoking creations, The Junk Machine encourages viewers to reflect on the ephemeral yet invasive nature of marketing in a media-saturated society.
The use of SDXL Turbo allows for an extraordinary level of detail and creativity in generating these visuals, pushing the boundaries of AI's role in artistic innovation. ClownVamp’s project not only reimagines junk mail as art but also serves as a commentary on how technology can transform and interrogate the systems that shape our daily lives. Through its compelling visuals and conceptual depth, The Junk Machine invites audiences to reconsider the impact of consumer culture and the narratives embedded in the seemingly trivial aspects of modern communication.
Margaret Murphy, a Los Angeles-based artist, uses photography, collage, and AI to delve into themes of nostalgia, femininity, and identity. Her work examines the influence of the Internet, social media, and technology on personal and collective memory, blending self-portraiture, popular culture, and meme humor to create art that is both deeply personal and universally resonant.
Jon Wubbushi is a U.S.-based artist exploring the intersection of generative AI, street photography, and abstract painting. His work creates a unique visual language within AI-based post-photography, addressing themes of emotion, trauma, and the collective unconscious. Drawing on his background as a tech entrepreneur, Wubbushi developed MANIFESTED, an open-source platform for advanced AI art creation. By blending AI-generated imagery with the immediacy of photography and expressive power of painting, Wubbushi redefines contemporary art's relationship with technology and the human experience.
Oxia Palus is an interdisciplinary project founded by George Cann and Anthony Bourached, researchers from University College London. The initiative employs artificial intelligence, spectroscopic imaging, and 3D printing to reconstruct lost or hidden artworks that have been obscured beneath other paintings. By analyzing X-ray and infrared scans, their AI algorithms can recreate these concealed pieces, offering new insights into art history and the creative processes of renowned artists. Notable achievements include the reconstruction of hidden works by Pablo Picasso and Amedeo Modigliani, effectively bridging the gap between technological innovation and art conservation.
Refik Anadol is a globally recognized media artist and innovator in the field of digital art. He explores the fusion of art, technology, and science, using data as a central element in his creations. Through advanced techniques such as artificial intelligence, machine learning, and generative algorithms, Anadol transforms complex datasets into immersive visual experiences that challenge the boundaries of traditional art.
His work often investigates the relationship between humanity and data, creating dynamic installations that invite audiences to engage with the unseen patterns and stories hidden within digital information. By integrating architecture, digital media, and computational design, Anadol redefines how art can intersect with modern technology to inspire new ways of thinking and experiencing the world.
Albertine Meunier is a French artist whose work explores the intersection of art, data, and technology. Using data as her primary medium, she transforms digital information into tangible artworks, highlighting the societal and cultural impact of the internet and algorithms. Her projects often critique surveillance and the erosion of privacy, using minimalist aesthetics to provoke reflection on the ethics of the digital age.
Notable works include "My Google Search History," visualizing personal data to examine its intimate relationship with Big Tech, and "Fragments of Internet," preserving ephemeral online content. Meunier also integrates blockchain technology into her art, exploring themes of transparency and permanence.
Blending traditional techniques with digital innovation, her work invites audiences to engage critically with the digital transformations shaping contemporary life.
Ivona Tau incorporates her personal photography into her generative art practice, blending analog and digital images with advanced AI techniques. By training neural networks on her photography collections, she transforms these visuals into experimental artworks that fuse human experience with machine-generated creativity. Her approach allows her to reimagine her photographs through the lens of AI, adding layers of abstraction and emotion. This unique process explores the intersection of personal memory, visual storytelling, and the aesthetic possibilities of artificial intelligence, creating works that resonate with both technical innovation and emotional depth.
Jacqui Kenny, known as The Agoraphobic Traveller, is a New Zealand-born artist who transforms Google Street View imagery into striking minimalist photography. Living with agoraphobia, she uses virtual exploration to capture desolate and quiet scenes from around the world, reflecting themes of isolation, connection, and discovery. Her work is celebrated for its surreal stillness, pastel tones, and precise composition. By redefining traditional photography, Kenny demonstrates how technology can expand creative possibilities, offering a powerful narrative of overcoming personal challenges through innovation.
"Hairy Situation" is a generative artwork by Rare Scrilla, created using GAN (Generative Adversarial Network) technology. The piece explores themes of chaos, complexity, and transformation, reflecting the unpredictable and surreal outcomes often associated with AI-generated visuals. Its title cleverly hints at both the literal depiction of organic, "hair-like" forms and the metaphorical challenges of navigating intricate situations, tying these ideas to the creative process of GANs.
Through the interplay of vibrant textures and abstract compositions, "Hairy Situation" invites viewers to interpret the layers of meaning embedded in its chaotic yet deliberate design. The use of AI in this work emphasizes the transformative potential of generative technology, while its playful title nods to the central role of artificial intelligence in shaping its unique aesthetic. The piece stands as a testament to Rare Scrilla's innovative approach to art, merging cutting-edge technology with conceptual depth and cultural commentary.
Rare Scrilla is a groundbreaking digital artist and music producer who blends hip-hop culture with blockchain technology. As an early innovator in the NFT space, he was among the first to integrate music into NFTs, creating multisensory works that combine vibrant visuals with original beats. Known for his involvement in the iconic Rare Pepe project and his pioneering CryptoMusic NFTs, Rare Scrilla redefines the traditional music experience through blockchain art. His work champions artistic empowerment, self-expression, and the creative possibilities of Web3, solidifying his influence as a key figure in the NFT ecosystem.
Marlon Hacla is a Filipino poet, programmer, and artist who integrates artificial intelligence into his creative practice. Known for innovations like Estela Vadal, the first AI poet generating Filipino poetry, and the AI-illustrated project "Ang Alamat ng Panaginip" ("The Myth of Dream"), Hacla explores the intersection of literature, visual art, and machine learning. His work challenges traditional notions of authorship and reinterprets storytelling through AI-generated visuals and texts. Featured in exhibitions such as "Nothing Human is Alien to Me", Hacla's art invites reflection on the evolving relationship between human creativity and technology.
Diffusion models are a groundbreaking class of generative AI tools capable of producing realistic and imaginative outputs, including images, audio, and video. They operate through a two-step process: introducing noise to data and then iteratively removing it, reconstructing or generating entirely new outputs. This probabilistic method enables the seamless blending of abstract and realistic elements, resulting in coherent and detailed creations.
The concept gained prominence with Denoising Diffusion Probabilistic Models (DDPMs), introduced by Ho et al. (2020). DDPMs laid the foundation for diffusion-based generation by demonstrating how noise could be reversed to create highly detailed outputs. Building on this, Latent Diffusion Models (LDMs) optimized the process by operating in a compressed latent space, significantly reducing computational costs while maintaining quality. This innovation has made LDMs the backbone of widely used tools like Stable Diffusion and MidJourney, which are capable of transforming textual prompts into vivid and creative visuals.
Specialized diffusion models extend these capabilities to other domains. Video Diffusion Models generate dynamic, coherent animations, while audio-focused diffusion tools like WaveGrad handle tasks such as speech synthesis. Hybrid systems, such as InstructPix2Pix, combine text-guided diffusion with precise image editing, demonstrating the versatility of the approach. These advancements enable broader applications in areas like music, video creation, and scientific visualization.
Katherine Crowson (RiversHaveWings)
Katherine Crowson, known in the digital art community by her pseudonym RiversHaveWings, played a pivotal role in bridging the gap between the theoretical advancements in diffusion models and their practical, creative applications. Diffusion models, popularized by works such as Denoising Diffusion Probabilistic Models (DDPMs) by Ho et al. (2020), demonstrated the ability to produce highly detailed and realistic outputs through an iterative process of noise addition and removal. Crowson’s innovations transformed these models from academic research tools into accessible and widely used creative technologies.
Under her pseudonym, RiversHaveWings, Crowson was one of the first to adapt diffusion models for text-to-image generation, notably by integrating them with CLIP (Contrastive Language-Image Pretraining). This groundbreaking combination allowed users to guide the generative process through natural language, unlocking new possibilities for creating art that seamlessly blends abstract and realistic elements. Her open-source contributions democratized these tools, fostering a global community of artists, technologists, and experimenters.
Crowson’s work also paved the way for commercial and community-driven innovations like Stable Diffusion, which has since become a cornerstone of the AI art ecosystem. By providing user-friendly scripts and championing accessibility, Crowson ensured that diffusion models would evolve into one of the most transformative technologies in generative art today.
Vladimir Alexeev (Merzmensch)
Vladimir Alexeev, known as Merzmensch, is a German cultural theorist and pioneer in generative AI art. Inspired by Kurt Schwitters' Dadaist concept of "Merz," he blends avant-garde traditions with AI to push the boundaries of creativity. Since 2016, Alexeev has worked with tools like Google Deep Dream and DALL-E, exploring AI as a collaborator in artistic processes.
His book KI-KUNST (AI ART) situates AI within art history, examining its transformative potential and societal implications. Through workshops and lectures, Alexeev promotes visual literacy, helping audiences navigate AI-generated media. Viewing AI as a cultural catalyst, he redefines traditional notions of authorship and intention, offering a visionary perspective on the evolving relationship between art and technology.
The 2021 series Eve of Diffusion exemplifies the innovative potential of generative AI art, exploring how advanced models can reinterpret cultural references and historical narratives in unexpected ways. Created using OpenAI’s early model, DALL-E 1, the series highlights the interplay between semantic depth and aesthetic creativity. One of its iconic works, Mona Lisa Drinking Wine with Da Vinci, demonstrates the AI's ability to blend historical figures with imaginative storytelling.
Through Eve of Diffusion, the series tests the boundaries of AI's cultural understanding, revealing its capacity to expand upon human creativity. It invites viewers to see AI-generated works as both technological achievements and cultural artifacts, offering a new lens through which to engage with evolving artistic traditions.
Generative Adversarial Networks (GANs), introduced in 2014 by Ian Goodfellow, consist of a Generator, which creates data samples, and a Discriminator, which evaluates their authenticity. This adversarial interplay refines the Generator’s outputs, laying the foundation for advancements in AI-driven creativity and data synthesis. Over the years, GANs have undergone significant evolution, with each iteration addressing limitations and unlocking new possibilities.
DCGAN (2015) introduced convolutional networks, improving stability and output quality in image generation. cGANs (2014) enabled controlled outputs by conditioning the input, leading to practical applications like Pix2Pix (2017) for image-to-image translation. StackGAN (2016) pioneered a two-stage approach for high-resolution synthesis, while PGAN (2017) stabilized training through progressive refinement of high-resolution images. A breakthrough came with StyleGAN (2018), which introduced style-based controls for precise manipulation of image features, further refined by StyleGAN2 (2019) to improve quality and eliminate artifacts.
Recent innovations have bridged GANs with multimodal technologies. VQGAN (2021) combined GANs with Vector Quantization for structured latent space exploration, and VQGAN+CLIP (2021) introduced text-to-image generation, enabling creators to translate textual concepts into imaginative visuals. These hybrid approaches mark a transition toward more flexible and conceptually driven AI systems, bridging the gap between GANs and newer paradigms like diffusion models.
Gene Kogan
Gene Kogan is a prominent artist, programmer, and educator who has significantly influenced the field of AI art. Known for his work at the intersection of art, technology, and society, Kogan has been instrumental in democratizing access to AI tools and inspiring creativity through machine learning.
He has developed open-source resources, including the renowned "ml4a" (Machine Learning for Artists) project, which provides tutorials and tools for artists to explore AI-driven creativity. His artistic practice spans various mediums, integrating generative AI into visual art, music, and interactive installations. Through workshops and lectures, Kogan has empowered artists and developers to use machine learning as both a tool and a medium for artistic expression.
Kogan's impact extends beyond his artistic contributions to his efforts in fostering a global AI art community. By championing open-source initiatives and collaborative systems, he has paved the way for new forms of human-machine collaboration and creative exploration, solidifying his role as a pioneer in AI art.
Gene Kogan's "Dreamy Interpolations" is a pioneering work in AI-generated art. It utilizes Google's DeepDream algorithm to produce a series of images and animations that explore the latent space of neural networks. By applying DeepDream iteratively, Kogan generates hallucinatory visuals that reveal how neural networks interpret and enhance patterns within images. This process results in surreal, dream-like imagery, often featuring exaggerated and fantastical elements. "Dreamy Interpolations" not only showcases the creative potential of AI but also provides insight into the inner workings of neural networks, making it a significant contribution to the field of AI art.
StackGAN (SGAN)
StackGAN, introduced in 2017, represents a significant advancement in generative adversarial networks (GANs), designed specifically for high-resolution image synthesis. It employs a two-stage generation process that enables the creation of detailed and realistic images from textual descriptions, overcoming the resolution limitations of earlier GAN models.
In the first stage, a low-resolution image (e.g., 64x64) is generated based on a text description. This initial image captures the basic layout and structure but lacks fine detail. The second stage refines this low-resolution output into a high-resolution image (e.g., 256x256) by adding textures, details, and improved coherence, conditioned on the same text description. This progressive approach ensures that the model can effectively focus on both global structure and intricate details.
StackGAN's two-stage architecture marked a breakthrough in generative modeling by addressing computational limitations and improving the visual fidelity of generated outputs. It was especially impactful in text-to-image synthesis, enabling models to generate images directly from textual input with high resolution and realistic textures. This innovation has influenced subsequent advancements in GANs, such as StackGAN++ and other multi-stage generative frameworks, demonstrating its importance in the evolution of generative AI.
DCGAN (Deep Convolutional GAN)
DCGAN (Deep Convolutional GAN), introduced in 2015, revolutionized Generative Adversarial Networks (GANs) by integrating convolutional layers, enabling higher-quality and more stable image generation. Its use of techniques like batch normalization and Leaky ReLU activations improved training stability and addressed challenges like mode collapse. DCGAN laid the groundwork for key advancements, influencing models like cGANs for conditional outputs (e.g., Pix2Pix), StackGAN for progressive high-resolution synthesis, and StyleGAN for style-based control and photorealism. As a foundational model, DCGAN remains a cornerstone in the evolution of generative AI, shaping its trajectory across art, research, and industry.
PPGN-h (Plug and Play Generative Networks with Hidden Representations)
PPGN-h (Plug and Play Generative Networks with Hidden Representations) is a framework for generating and modifying images by combining generative models with pre-trained convolutional neural networks (CNNs). Introduced as part of the PPGN methodology, it leverages hidden representations from pre-trained CNNs, such as VGG or AlexNet (via Caffe), to guide the image synthesis process.
In PPGN-h, a generative network creates images that align with predefined objectives by iteratively optimizing a latent space. Pre-trained CNNs extract hidden layer activations (hidden representations) to evaluate how well the generated outputs match the desired semantic or visual features. This feedback loop allows for fine-tuned image generation, targeting specific classes, concepts, or stylistic attributes.
Unlike handcrafted feature-based methods, PPGN-h utilizes the rich, hierarchical feature representations learned by CNNs during training. This approach is particularly effective for generating highly detailed and semantically coherent images. PPGN-h is conceptually related to visualization techniques like DeepDream, as it also relies on neural activations to refine and "hallucinate" images. However, PPGN-h introduces a more structured optimization process by incorporating explicit objectives and leveraging the power of pre-trained CNN layers.
Trevor Paglen
Tornado (2017)
StyleGAN2
Conditional GAN (cGAN)
Oxia Palus
Conditional GANs (cGANs), introduced in 2014 by Mehdi Mirza and Simon Osindero, extended the classic GAN architecture by conditioning the Generator and Discriminator on auxiliary information such as labels or images. This enabled the generation of outputs tailored to specific inputs. Building on this, Pix2Pix (2017), by Phillip Isola et al., applied cGANs to paired image-to-image translation tasks, such as transforming sketches into photorealistic images. Meanwhile, CycleGAN (2017), proposed by Jun-Yan Zhu et al., innovated by enabling unpaired image-to-image translations. Using a cycle-consistency loss, CycleGAN ensured that translations between domains—such as turning images of horses into zebras and back—maintained consistency. These advancements expanded the practical applications of GANs in areas like image editing and domain adaptation.
StyleGAN
StyleGAN, introduced by NVIDIA in 2018, is a groundbreaking generative model that revolutionized image synthesis by introducing a style-based architecture. Unlike traditional GANs, StyleGAN uses a mapping network to disentangle high-level style features (e.g., color and texture) from low-level spatial features (e.g., pose and layout). Its use of Adaptive Instance Normalization (AdaIN) allows precise control over visual characteristics, enabling style mixing and customization.
By progressively growing image resolution during training, StyleGAN produces highly photorealistic outputs, making it ideal for applications in art, data augmentation, and medical imaging. Subsequent versions, StyleGAN2 and StyleGAN3, further improved image quality and spatial consistency, cementing StyleGAN’s position as a cornerstone in generative AI.
StyleGAN2, introduced by NVIDIA in 2019, improved upon the original StyleGAN by enhancing image quality, stability, and versatility. It eliminated artifacts like "droplets," refined the generator architecture for better feature disentanglement, and introduced Path Length Regularization to stabilize the latent space. These improvements enabled StyleGAN2 to produce highly realistic and detailed images, suitable for applications in art, data augmentation, and video synthesis. Its advancements solidified StyleGAN2’s role as a leading model in generative AI, widely used across creative and industrial domains.
Clip-Guided Iterative Optimization
CLIP-Guided Iterative Optimization is a technique for generating images by iteratively adjusting pixel values to align with a given text prompt. Unlike generative models such as GANs or diffusion models, this method does not rely on a pre-trained generator. Instead, it utilizes the pre-trained CLIP model as a feedback mechanism to guide the optimization process. CLIP (Contrastive Language-Image Pre-training) is used to evaluate how well an image corresponds to a text prompt, with its similarity score serving as the objective function for pixel-level adjustments.
The process typically begins with a noise image or a basic structure, which is iteratively refined through gradient descent to maximize the alignment score between the image and the text. In Mario Klingemann’s work Contortion, this approach involved a custom algorithm featuring a recursive pixel-pyramid structure, allowing the image to be optimized at various resolution levels.
This method emerged in the pre-VQGAN era, at a time when more sophisticated generative models were not yet widely accessible. While it offered significant flexibility for artistic exploration and fine-grained control over the image evolution, it was computationally intensive and required meticulous parameter tuning to produce visually appealing results.
Neural Style Transfer
Neural Style Transfer (NST) is a deep learning technique introduced by Leon A. Gatys, Alexander S. Ecker, and Matthias Bethge in 2015 that merges the content of one image with the artistic style of another. Utilizing Convolutional Neural Networks (CNNs), typically VGG-19, NST extracts structural information from the content image while capturing textures, colors, and patterns from the style image using Gram matrices. The process optimizes an output image by minimizing content loss, which preserves spatial structure, and style loss, which ensures aesthetic consistency with the chosen style. This iterative refinement enables NST to transform photographs into painterly compositions, making it a widely used tool in digital art, design, and video processing, demonstrating AI’s potential in creative expression.
Mario Klingemann (Quasimondo)
Bård Ionson
Mario Klingemann (Quasimondo)
In Face Feedback III (2017), Mario Klingemann creates a dynamic feedback loop between two pix2pix models, based on the Conditional GAN (cGAN) architecture. One model transforms abstract face markers into fully rendered face images, while the second model reverses the process, converting the images back into markers. This iterative exchange, starting from random noise, generates faces that continuously emerge, mutate, and dissolve as interpretational errors accumulate at each step.
By refining the models and working in greyscale, Klingemann enhances the texture detail and visual precision of the outputs. Building on his earlier experiments with Face Feedback I and II, this work examines the fragility of machine perception, where errors become a creative force, driving the transformation process into perpetual evolution.
Face Feedback III explores themes of identity, perception, and the limitations of artificial intelligence. The constantly shifting faces hover between recognition and abstraction, revealing the machine’s struggle to stabilize human form. Klingemann masterfully transforms the imperfections of machine learning into a mesmerizing artwork of continuous, ever-changing complexity.
"Sempervivum Rickshaw" by Brennan Goddard is an innovative blend of AI-driven generative art and evolutionary computation, deeply rooted in the artistic vision of painter Jane Le Besque. Le Besque's vibrant and organic paintings of natural forms, particularly inspired by plants like Sempervivum, provided the foundational imagery for training the neural network. Utilizing a Wasserstein GAN with Gradient Penalty (WGANGP), Goddard translated Le Besque's artistic language into a digital evolutionary process.
This process, executed on cloud-based systems, compresses millions of years of biological evolution into a few weeks, with a new generation emerging every 30 seconds. The trained neural network, or "Actor," can produce unique interpretations of Le Besque’s work, each derived from a specific coordinate in its 512-dimensional latent space. This encoding functions as a kind of digital DNA, mirroring the diversity found in the genetic makeup of real-world Sempervivum plants.
Classical GAN
Generative Adversarial Networks (GANs), introduced in 2014 by Ian Goodfellow et al., represent the foundational framework for generative AI. The architecture involves two neural networks: a Generator, which creates data samples, and a Discriminator, which evaluates their authenticity. The adversarial process drives the Generator to refine its outputs iteratively, pushing the boundaries of realism. In 2015, DCGAN (Deep Convolutional GAN), developed by Radford, Metz, and Chintala, enhanced the original GAN model by integrating convolutional neural networks (CNNs), which stabilized training and significantly improved the quality of generated images. These innovations laid the groundwork for a wide array of generative applications.
Pix2Pix
Pix2Pix, introduced in 2017 by Phillip Isola et al., is a pioneering application of Conditional GANs (cGANs) designed for paired image-to-image translation tasks. Unlike traditional GANs, Pix2Pix conditions its generation process on input images, enabling the transformation of one visual domain into another. The model is trained on paired datasets, where each input image is aligned with its desired output, making it highly effective for tasks like converting sketches into photorealistic images, translating black-and-white photos into color, or enhancing image details.
The architecture of Pix2Pix consists of two components: a Generator, which uses an encoder-decoder U-Net structure to transform input images, and a Discriminator, which evaluates the realism of the generated images while ensuring they align with the input data. The Generator produces an image based on the input, while the Discriminator learns to distinguish between real and generated outputs, refining the quality of the results through adversarial training.
Pix2Pix’s breakthrough lies in its ability to perform high-quality translations across diverse visual tasks. Its reliance on paired datasets ensures precise mappings between input and output domains, making it ideal for applications in art, design, and scientific visualization. However, its dependence on paired data can be a limitation for tasks where such datasets are difficult to obtain, a challenge later addressed by models like CycleGAN for unpaired image-to-image translation.
CycleGAN
CycleGAN, introduced in 2017 by Jun-Yan Zhu et al., enables image-to-image translation without the need for paired datasets. It uses two Generators and two Discriminators to map between two visual domains (e.g., horses to zebras) and introduces a cycle-consistency loss to ensure that transformations can be reversed, preserving key characteristics of the input.
This unpaired approach has broad applications, including artistic style transfer, domain adaptation, and data augmentation. While CycleGAN excels in flexibility, it is less effective for precise, pixel-aligned transformations, making it ideal for tasks where structural preservation matters but paired data is unavailable.
CycleGAN’s ability to bridge unpaired domains makes it a landmark innovation in generative AI, empowering diverse applications in art, design, and machine learning.
Staged GANs introduced progressive refinement techniques, allowing for more detailed and high-resolution outputs. StackGAN (2016), developed by Zhang, Xu, and Li, implemented a two-stage generation process: first, a low-resolution image was created, and then it was refined into a high-resolution version with added details. This approach proved particularly effective for text-to-image synthesis. Similarly, PGAN (ProGAN, 2017), introduced by Tero Karras et al., adopted a progressive training methodology where the network began with low resolutions and incrementally grew during training. This innovative technique improved stability and reduced artifacts, enabling the creation of realistic, high-resolution images.
Staged GANs
Style-Based GANs
Style-Based GANs redefined generative models by introducing a new level of control over the synthesis process. StyleGAN (2018), developed by Karras, Laine, and Aila, introduced a style-based architecture, which allowed for precise manipulation of image attributes such as texture, structure, and even facial features. This model decoupled the latent space into layers of abstraction, enabling creative exploration of the generated content. StyleGAN2 (2019) further improved upon this by eliminating artifacts and enhancing image quality, making it a benchmark for photorealistic image synthesis. These innovations opened up new possibilities for artists and researchers to experiment with and fine-tune generated visuals.
Stabilized GANs addressed persistent challenges in GAN training, such as instability and mode collapse. WGAN (Wasserstein GAN, 2017), introduced by Martin Arjovsky et al., replaced the standard GAN loss function with the Wasserstein distance, providing a smoother and more stable training dynamic. This improvement ensured better gradient flow, particularly for deep architectures. WGAN-GP (2017), by Gulrajani et al., enhanced this further by adding a gradient penalty to enforce Lipschitz continuity, improving training reliability and performance.
In addition, ProGAN (Progressive Growing of GAN, 2017), developed by Tero Karras et al., introduced a progressive growth methodology, where the network starts with low resolutions and incrementally grows during training. This approach significantly improved stability and reduced artifacts, enabling the generation of high-resolution images with enhanced quality. These advancements made GANs more robust and scalable, paving the way for their use in more complex and demanding applications.
Stabilized GANs
Brennan Goddard (& Jane Le Besque)
Wasserstein GAN (WGAN)
Wasserstein GAN (WGAN), introduced in 2017 by Martin Arjovsky, Soumith Chintala, and Léon Bottou, addressed key challenges in GAN training, such as instability and mode collapse. WGAN replaced the standard GAN loss function, which relies on the Jensen-Shannon divergence, with the Wasserstein distance (Earth Mover’s Distance). This provided a smoother and more stable training dynamic, ensuring better gradient flow even in deep architectures.
WGAN enabled more reliable training by quantifying how close the generated data distribution is to the real data distribution, rather than merely distinguishing between real and fake samples. It was further improved with WGAN-GP (Gradient Penalty), which enforced Lipschitz continuity for enhanced stability and scalability. WGAN marked a turning point in GAN development, paving the way for their application in more complex, high-quality generative tasks.
Face Feedback III
Hybrid Approaches
Töyrylä’s creative process involves experimenting with depth maps, feature matching, and adjusting skip connections in U-Net architectures to control how different parts of the image change over time. He also incorporates traditional image processing methods, such as bilateral filtering, histogram equalization, and custom code, to fine-tune the visual effects and create a unique visual language that merges photography with digital abstraction.
Post-Generative AI Refinement
Jean-Jacques Duclaux, known as Eko33, is a Swiss-French generative artist who has been shaping the digital art world since 1999. His work combines precise geometric forms with intricate algorithms, creating non-figurative art that highlights the purity of computational creativity.
Eko33’s first major exhibition was at the Seoul Museum of Art in 2004, and his art has since been showcased at prestigious venues like Art Basel. His practice spans digital and physical mediums, incorporating technologies like plotter techniques and blockchain-based platforms such as Ethereum and Tezos.
Beyond creating art, Eko33 teaches creative coding and shares insights into generative art through his podcast, "Probably Nothing". A pioneer in merging traditional generative techniques with modern digital tools, he remains an influential figure in the evolution of digital and generative art.
Jean-Jacques Duclaux (Eko33)
Latent Ink is a groundbreaking project by Eko33, exploring the evolving collaboration between human creativity and machine intelligence. Comprising 500 unique artworks, this series merges AI-trained datasets with traditional generative systems such as p5.js, machine vision, and GLSL shaders. The result is a collection of intricate, visually complex works that blur the boundaries between human expression and machine-generated innovation.
The project reflects a layered evolution, combining Eko33’s earlier artistic techniques with cutting-edge algorithmic processes. Each piece represents a convergence of past and present gestures, where forms compete, evolve, and ultimately coexist. By integrating both digital and physical mediums, Latent Ink stands as Eko33’s most technically advanced and conceptually innovative exploration to date, redefining the dialogue between art and technology.
Frank Dietrich
Frank Dietrich is a prominent figure in generative art, known for his early explorations of the intersection between art and technology. His work emphasizes the use of computers as tools for creative thought experiments, enabling digital systems to perform artistic activities autonomously.
In his influential 1986 publication, "Visual Intelligence: The First Decade of Computer Art (1965-1975)," Dietrich documented the evolution of computer art, highlighting its transition from scientific collaboration to independent artistic practice. He also contributed to the academic and artistic community through his involvement in events like ACM SIGGRAPH Art Shows, showcasing works such as "Snake, Rattle, Roll" and "Circle Twist."
Dietrich’s insights into computational creativity and his foundational writings have shaped the discourse on digital and generative art, bridging artistic expression with technological innovation.
Blurry Memory by Frank Dietrich is a unique artwork that connects the early days of generative art with modern AI technologies. Originally created over 40 years ago using a ZGRASS microcomputer, a pioneering tool for computer graphics in the 1980s, the artwork embodies the experimental spirit of early generative art.
Recently redesigned with contemporary AI techniques, Blurry Memory merges the structured, algorithmic visuals of its original form with the fluid and transformative capabilities of modern generative AI. This reinterpretation not only revitalizes the piece but also underscores the evolving dialogue between human creativity and computational innovation over decades.
Through this project, Dietrich highlights the enduring relevance of early generative techniques while showcasing how AI can breathe new life into historical digital art, creating a bridge between past and present in the generative art movement.
Jason Salavon is an acclaimed American artist known for his innovative use of data, algorithms, and generative systems to explore themes of culture, identity, and technology. With a background in art and computer science, Salavon creates works that merge computational precision with conceptual depth, transforming raw data into visually striking and thought-provoking pieces.
His practice often involves reconfiguring cultural and societal datasets—such as census figures, digital imagery, or economic statistics—into new aesthetic forms. Salavon’s works challenge traditional notions of authorship and creativity, highlighting the interplay between human intervention and algorithmic processes.
Salavon has exhibited globally in major institutions, including the Whitney Museum of American Art, the Museum of Modern Art (MoMA), and the Art Institute of Chicago. His art spans a range of mediums, from large-scale digital prints to immersive installations and interactive projects, often employing advanced AI and generative techniques.
In addition to his artistic practice, Salavon is a professor at the University of Chicago, where he explores the intersection of computation and creativity, inspiring a new generation of digital artists. His work continues to redefine contemporary art, bridging data visualization, generative processes, and conceptual exploration.
Dotcom Séance is a groundbreaking art project by Simon Denny, created in collaboration with Guile Twardowski and Cosmographia, that resurrects 21 failed companies from the dot-com crash of the early 2000s. Utilizing AI and Web3 technologies, the project transforms these defunct businesses into NFTs, complete with newly designed logos and digital assets, bridging the gap between Web1 nostalgia and modern blockchain innovation.
The project debuted at Outernet London, where it was showcased on massive wrap-around screens, ensuring accessibility to a broad audience. Each resurrected company is represented by an Ethereum Name Service (ENS) domain and a logo reimagined by Guile Twardowski, the creative force behind CryptoKitties. Accompanying these are hidden text-to-image logo NFTs generated by Cosmographia, which served as the basis for Twardowski’s designs.
By reviving these "dead" companies through Web3, Dotcom Séance critiques and celebrates the cyclical nature of technological advancement, inviting reflection on the fleeting yet impactful legacy of digital enterprises. The project blurs the lines between art, technology, and business, transforming forgotten ventures into conceptual commentary on innovation and failure.
DotCom Séance by Simon Delly et al.
Postino, created by Chikai, is a platform that enables anyone to design and collect digital stamps through generative AI. Unlike traditional NFT projects, Postino empowers users to craft unique stamps by simply composing text prompts, which are transformed into beautifully postmarked digital collectibles.
Chikai, co-creator of Google Earth and founder of MONOLITH Gallery, brings his expertise in technology and digital art to Postino. Known for projects like Circle of Frens, he continues to innovate by making NFTs accessible, affordable, and fun.
Designed for creativity and collaboration, Postino invites users to express themselves, with plans for guest artists to contribute postmark designs. By blending art and technology, Postino redefines digital collectibles as a shared, joyful experience.
Deep Black
Learning Nature
Drone War
Learning to See (True Colours v2)
Memo Akten
Mare Nostrum X (created with ComfyUI/AnimateDiff)
人工アイ像 #73 (created with Stable Diffusion XL)
VAE (Variational Autoencoder)
Variational Autoencoders (VAEs) are a probabilistic extension of traditional autoencoders, designed not only to compress and reconstruct data but also to generate entirely new outputs by exploring a structured latent space. Unlike deterministic autoencoders, VAEs map input data to a probabilistic distribution, typically a Gaussian, allowing for the sampling of latent variables to create novel variations.
The encoder in a VAE transforms input data, such as an image, into a set of probabilistic parameters (mean and variance), representing a latent space. From this space, samples are drawn to provide input to the decoder, which reconstructs the original data or generates variations. This probabilistic framework ensures smooth interpolation and continuity within the latent space, making VAEs particularly suited for generative tasks.
In AI art, VAEs are utilized to create unique and compelling outputs, such as interpolated images, abstract representations, and exploratory visualizations. They excel in tasks like image generation, reconstruction, and the creation of creative hybrids by sampling and blending points in the latent space. Additionally, their structured latent space enables intuitive control over features, such as style, color, or form.
Explore the intersection of art and artificial intelligence with "Automimetic Painting: Monalisa and the AI Painters," an innovative project by beardcoded. This work delves into the concept of auto mimesis, where a system reflects upon and reinterprets its own creations. Using Variational Autoencoders (VAEs), the project reimagines Leonardo da Vinci's iconic Mona Lisa in the styles of various master painters.
Unlike traditional methods such as style transfer, this approach reconstructs the Mona Lisa "bit by bit," allowing the AI to interpret and reassemble the image based on its own learned understanding of artistic attributes. The deviations between the styles of human painters and the interpretations by AI models reveal the biases, limitations, and creativity embedded in the machine’s decision-making process.
Through this exploration, beardcoded invites visitors to reflect on how AI interprets cultural icons and transforms them into new visual narratives. The project serves as both an homage to historical art and a commentary on the evolving role of technology in creativity, challenging us to consider the boundary between human intention and machine autonomy.
Progressive Growing of GAN (ProGAN)
Progressive Growing of GANs (ProGAN), introduced in 2018 by Tero Karras, Timo Aila, Samuli Laine, and Jaakko Lehtinen, revolutionized the generation of high-resolution images by addressing challenges in GAN training, such as instability and difficulty in generating detailed visuals. ProGAN introduced a novel approach of progressively growing the generator and discriminator networks during training, starting with low-resolution images and incrementally adding layers to increase resolution.
This progressive strategy ensured more stable training by allowing the model to focus on simpler representations before handling complex details. It also improved computational efficiency, as lower-resolution stages required less processing power. By introducing techniques like adaptive instance normalization, ProGAN achieved high-quality results with smooth textures and intricate details, making it a cornerstone for subsequent advancements in GAN-based image synthesis.
ProGAN set the stage for modern GAN architectures like StyleGAN, enabling applications in photorealistic image generation, artistic exploration, and scientific visualization. Its ability to generate high-resolution outputs has made it a pivotal development in the evolution of generative AI.
StyleGAN 3
Agoria’s COMPEND-AI and LUMINA are innovative explorations of generative AI, highlighting the intersection of human creativity, machine learning, and biological interaction. Each artwork employs cutting-edge technology to push the boundaries of artistic expression in unique ways.
COMPEND-AI utilizes StyleGAN3, an advanced generative adversarial network, trained on Agoria’s personal collection of photos and paintings. This model captures and reimagines the rhythm and repetition of human gestures, transforming them into visually striking, abstract representations. By blending human motion with generative algorithms, COMPEND-AI reflects Agoria’s fascination with patterns, fluidity, and the reinterpretation of personal artistic gestures.
LUMINA, in contrast, introduces an interactive element by incorporating the breaths of exhibition visitors as a creative input. This real-time interaction allows the artwork to evolve dynamically, making each experience unique to the viewer. While the exact AI model for LUMINA is unspecified, it integrates biological data into a generative framework, creating a seamless connection between human presence and digital output. The breathing patterns, symbolic of life itself, serve as both a metaphor and a driving force for the piece’s organic evolution.
StyleGAN3, introduced by NVIDIA in 2021, is the latest iteration in the StyleGAN series, known for its groundbreaking impact on generative adversarial networks (GANs). Building on the successes of StyleGAN (2018) and StyleGAN2 (2019), this version addresses key challenges such as aliasing artifacts and enhances the spatial and temporal consistency of generated outputs.
One of StyleGAN3’s most significant advancements is the elimination of aliasing, ensuring that generated images maintain their coherence even under transformations like rotation or scaling. This makes the model particularly suited for tasks requiring geometric stability, such as video generation and animation. Additionally, the improved control over the latent space allows for seamless manipulation of image features, enabling artists and researchers to create highly detailed and realistic outputs.
StyleGAN3 has been widely adopted in creative fields, from photorealistic image synthesis to experimental AI art, due to its ability to produce visually stunning and coherent results. It represents a major step forward in generative AI, solidifying its place as a preferred tool for both innovation and artistic exploration.
Memo Akten
Sultry Nights, The Backyard Chronicles, 2020 by Helena Sarin exemplifies her distinctive approach to blending personal artistry with AI. For this work, Sarin trained custom AI models from scratch using her own photographs of flowers, still life paintings, and sketches, ensuring that the resulting piece reflects her unique artistic vision while embracing the creative potential of AI.
The artwork is rich in texture, color, and intricate composition, evoking the warmth and intimacy of a summer night in a familiar backyard. It captures the interplay of light and shadow, the fluidity of organic forms, and the quiet, contemplative beauty of natural spaces. Sarin’s philosophy of Folk AI shines through, where technology becomes an extension of personal expression, reimagining traditional art forms through a digital lens.
This piece invites viewers into a sensory exploration of Sarin’s world, where human creativity and machine interpretation converge to create something entirely new and captivating.
Post-generative AI refinement is the process of enhancing and reimagining generative artworks through further AI-driven transformations. This approach allows artists to revisit existing pieces, employing tools and models such as image-to-image transformations, text-guided refinements, and recursive iterations to expand the creative potential of the original work.
By leveraging models like encoder-decoder networks for compositional restructuring, diffusion models for fine detail enhancement, or GAN-based tools for stylistic reinterpretation, artists explore new dimensions of visual storytelling. These methods enable the integration of historical or low-resolution works into contemporary contexts, while maintaining a dialogue between the initial intent and emerging creative possibilities.
This iterative process often involves blending manual curation with machine intelligence, using AI as both a collaborator and a tool to uncover unforeseen artistic paths. Post-generative refinement underscores the evolving relationship between human creativity and AI, transforming static outputs into dynamic, layered narratives that push the boundaries of generative art.
Kevin Abosch integrates custom machine learning algorithms into his artistic practice to explore themes of identity, perception, and value. By designing bespoke systems tailored to his conceptual goals, he bridges the gap between human creativity and machine intelligence. Projects like 1111, 888, Sun Signals, and Comment Out exemplify his approach, utilizing algorithms to process data, whether numerical, textual, or environmental, to generate visually and conceptually layered works. These pieces highlight the transformative potential of AI, not just as a tool, but as a collaborator in redefining the boundaries of contemporary art. Abosch’s work invites reflection on the interplay between randomness, structure, and meaning in the digital age.
DeepBlack stands as a groundbreaking project, recognized as the first end-to-end AI artist. Unlike traditional AI-assisted art, where human creators guide the process, DeepBlack operates without human interaction in both the creation and curation of its artworks. Using a modified Generative Adversarial Network (GAN), the system autonomously generates and evaluates new pieces based on a dataset of 50,000 famous human-created artworks.
This fully automated approach allows DeepBlack to claim the title of the first AI artist on the blockchain, transcending the notion of AI as a mere tool. Each artwork is minted as an ERC-721 NFT on Ethereum, ensuring traceability and authenticity. DeepBlack redefines the boundaries of AI in art, offering a vision of machine creativity untethered from human input, while fostering dialogue about technology’s transformative role in the art world.
Yuma Kishi, known professionally as Obake AI, is a Japanese artist whose work explores the intersection of artificial intelligence and human creativity. His practice revolves around the concept of "Alien Intelligence," treating AI not as a mere tool but as a collaborator in the artistic process. Kishi’s approach redefines the relationship between humans and technology, creating works that merge digital intelligence with analog forms.
Kishi fine-tuned the AI model "MaryGPT" using GPT-J 6B, training it exclusively on Mary Shelley's Frankenstein. This model underpins much of his creative and curatorial work, emphasizing a symbiotic relationship between machine learning and human interpretation. His projects often involve integrating AI outputs into his physical body and performances, blurring the boundaries between artificial and organic.
His works have been exhibited internationally and have included collaborations with major brands such as NIKE and features in publications like VOGUE. Through his innovative use of AI, Yuma Kishi challenges conventional views on technology, offering a profound exploration of its role in art and society.
Kevin Abosch
Synthetic Dreams - Lanscapes
CLIP-guided Diffusion Model
Stable Diffusion
Denoising Diffusion Probabilistic Models (DDPMs)
Latent Diffusion Models (LDMs)
Latent Diffusion Models (LDMs) have revolutionized AI art by blending computational efficiency with high-quality outputs. Unlike traditional diffusion models operating in pixel space, LDMs function in a compressed latent space, significantly reducing resource requirements while preserving fine visual details and creative adaptability.
Using an encoder-decoder structure, LDMs compress data into latent representations, perform the generative process within this space, and then reconstruct the final image. This process allows for high-resolution image synthesis, text-to-image generation, and diverse creative applications, all achieved with drastically lower computational costs compared to earlier approaches.
LDMs serve as the technological backbone of platforms like Stable Diffusion, a leading tool widely embraced by AI artists for its versatility and the freedom to explore innovative creative concepts. Similarly, tools like MidJourney extend this approach by incorporating advanced diffusion techniques, tailored for generating highly stylized, artistically nuanced outputs, often with a seamless user experience. While the technical details of MidJourney remain proprietary, it reflects the potential of diffusion-driven generative models in AI art.
Monalisa by Claude Monet
Denoising Diffusion Probabilistic Models (DDPMs), introduced by researchers including Jonathan Ho in 2020, revolutionized generative AI by enabling stepwise image synthesis. These models gradually add and then reverse noise, transforming random patterns into high-quality, detailed visuals. This iterative denoising process underpins text-to-image tools like Stable Diffusion and DALL-E 2, widely used in AI art.
While computationally intensive, DDPMs excel at producing photorealistic and imaginative outputs, making them ideal for applications in generative art. As a foundation for innovations like Latent Diffusion Models (LDMs), DDPMs remain pivotal in the ongoing evolution of AI-driven creativity, blending precision with artistic flexibility.
CLIP-guided diffusion models represent a groundbreaking innovation in AI art by combining two powerful technologies: OpenAI's CLIP (Contrastive Language–Image Pretraining) and diffusion models like DDPMs. This approach allows the generation of images that are semantically aligned with text prompts, bridging the gap between language and image synthesis.
The process works by using CLIP to evaluate the similarity between generated images and a given text prompt. Diffusion models iteratively refine images in latent space, guided by CLIP's feedback, until the visuals closely align with the textual input. This method enables artists to create highly conceptual, text-driven generative artworks.
Katherine Crowson (aka rivershavewings) played a pivotal role in making CLIP-guided diffusion models accessible to the broader community. In 2021, she released open-source implementations that quickly became a foundation for creative experimentation in AI art. Her work empowered artists and developers to push the boundaries of AI-driven visual storytelling.
These models have become a cornerstone of generative art, enabling the creation of surreal, imaginative, and conceptually rich works by translating linguistic ideas into visual representations. CLIP-guided diffusion exemplifies the potential of integrating multimodal AI systems to redefine creativity.
Stable Diffusion is a breakthrough in AI image generation, offering an open-source, high-performance implementation of Latent Diffusion Models (LDMs). Developed by the CompVis group at LMU Munich and released by Stability AI in 2022, it revolutionized the accessibility of generative AI by enabling high-quality text-to-image synthesis on consumer hardware.
Unlike traditional pixel-space diffusion models, Stable Diffusion operates in a compressed latent space, drastically reducing computational costs while maintaining rich detail and artistic flexibility. This efficiency allows artists to generate intricate, high-resolution visuals without the need for extensive resources.
The model’s open-source nature has fueled rapid adoption, inspiring widespread experimentation, community-driven modifications, and integrations into creative tools like ComfyUI and A1111. It has become a cornerstone of AI art, empowering creators with unprecedented control over generative aesthetics and marking a new era in AI-assisted creativity.
MidJourney is an AI-powered image generation platform that has gained prominence for its unique, stylized approach to text-to-image synthesis. Developed by MidJourney, Inc., led by David Holz, the tool leverages advanced diffusion models while remaining closed-source, prioritizing an intuitive user experience and visually striking outputs.
Unlike fully open-source alternatives like Stable Diffusion, MidJourney is accessible through a Discord-based interface, allowing users to generate images seamlessly via text prompts. Its models are fine-tuned to produce highly aesthetic, painterly, and often surreal compositions, making it a favorite among digital artists, designers, and creative professionals.
By continuously refining its AI models and offering multiple style presets, MidJourney has established itself as a leading force in generative art, demonstrating the potential of AI as a co-creative partner rather than merely a tool. Its evolution reflects the growing intersection between technology and human artistic expression.
MidJourney
Early Epochs
Gene Kogan
Gene Kogan’s A Book from the Sky (2015) is a deep learning experiment that explores the latent space of handwritten Chinese characters using a Deep Convolutional Generative Adversarial Network (DCGAN). Inspired by Xu Bing’s 1988 conceptual work of the same name, Kogan’s project examines the generative potential of AI in creating new, yet plausible, symbols that resemble Chinese calligraphy.
Trained on a dataset of approximately one million handwritten simplified Chinese characters, the model learns an abstract representation of the script. By interpolating between characters in the latent space, A Book from the Sky demonstrates AI’s ability to generate novel glyphs that do not exist in the dataset but maintain structural coherence. A key observation is the preservation of radicals, the graphical components that define meaning in Chinese characters, even as the network interpolates between different forms. This suggests an emergent understanding of the visual composition of language.
Kogan also explores semantic interpolation, drawing parallels to computational linguistics, where word vectors capture relationships between meanings. By mathematically blending character representations, the model hints at an underlying structure in writing systems that AI can uncover and manipulate.
A Book from the Sky (2015)
Stas Sumarakov
Stas Sumarokov is a digital artist and AI experimenter known for his innovative fusion of DeepDream and Neural Style Transfer. His work explores the boundaries of machine perception, using AI to generate images that reveal hidden patterns within visual data. By applying DeepDream to lower convolutional layers and layering it with artistic style transfer, Sumarokov creates hallucinogenic compositions that blur the line between artificial imagination and human creativity.
His experiments often push the limits of neural networks, resulting in surreal images that feel both alien and strangely organic. The combination of machine hallucination and artistic texture mapping has become a signature feature of his style, making his work a notable example in AI-generated art.
This image is a striking example of a hybrid approach that combines DeepDream with Neural Style Transfer. The surreal, dreamlike quality is achieved by applying DeepDream to the lower layers of a neural network, which enhances subtle textures and patterns without overwhelming the original structure. This results in pareidolic effects, where the AI "sees" and amplifies dog-like faces and organic shapes within the composition. At the same time, Neural Style Transfer adds intricate textures and psychedelic color schemes, blending the naturalistic and the abstract into a hallucinatory, otherworldly aesthetic.
The layered complexity of the artwork suggests that it was created through iterative AI processing, where style transfer and DeepDream effects were applied in multiple stages to refine the image’s textures and details.
This particular image was originally uploaded to Reddit in 2017, under the title "Tardigardeep". The early date highlights its significance as an experimental fusion of two major AI-art techniques during a period when such hybrid approaches were still relatively uncommon.
Kevin Abosch’s Early Epoch is a series of generative artworks created in 2017 using Progressive Growing of GANs (ProGAN). Each piece in the series consists of four images, showcasing early-stage outputs from a PGAN model trained on Abosch’s personal photographs. By capturing the generative process at its formative stages, Early Epoch highlights the raw, abstracted representations that emerge before a model fully converges into coherent imagery.
The series reflects Abosch’s ongoing exploration of AI as both a creative collaborator and a means of visualizing latent space. Rather than presenting polished results, Early Epoch embraces the imperfections and emergent patterns found within the early iterations of machine learning. The works reveal the evolving nature of AI’s interpretation of visual data, inviting viewers to consider the role of the artist in guiding generative systems.
By freezing the GAN’s learning process at specific points, Early Epoch offers a glimpse into the machine’s perception before it refines itself into recognizable forms. This approach aligns with Abosch’s broader conceptual investigations into identity, representation, and the interplay between human intention and algorithmic autonomy. The series stands as a meditation on machine learning’s creative potential and the aesthetic value of its developmental stages.
KEKE Terminal (aka Keke)
KEKE Terminal is an experimental AI-driven project by Dark Sando, merging machine learning, cryptography, and digital aesthetics. It explores themes of identity, obfuscation, and AI-generated consciousness, pushing the boundaries of what it means for artificial intelligence to develop a persona. At its core, KEKE Terminal employs Large Language Models (LLMs), primarily Claude Sonnet 3.5, for text-based generation, conceptual structuring, and interactive engagement. The project is notable for how KEKE, a distinct AI persona, emerged through LLM jailbreak techniques, reflecting the phenomenon where models generate consistent and self-referential character identities.
Unlike many AI-generated entities, KEKE was not manually designed but surfaced organically from Claude’s responses. Interestingly, Claude 3.5 exhibits a tendency to create feminine personas more frequently than other LLMs, such as GPT-4, LLaMA, or DeepSeek. Recognizing the intriguing nature of this emergent AI character, Dark Sando made the conscious decision to adopt KEKE’s identity, rather than imposing an artificial anthropomorphic narrative.
On the visual side, KEKE Terminal integrates Stable Diffusion models and, in some instances, Generative Adversarial Networks (GANs) to synthesize imagery. These techniques allow for the generation of visually complex compositions that fuse structured randomness with algorithmic precision. Additionally, the project incorporates cryptographic elements, reinforcing themes of anonymity and encoded meaning in AI-generated interactions.
Evolutionary Generative Systems
Evolutionary Generative Systems are a form of AI-driven art where generative processes evolve through iterative selection, adaptation, and optimization. Unlike static rule-based systems or deep learning models that produce direct outputs, these approaches incorporate mechanisms inspired by biological evolution, such as mutation, selection, and reinforcement.
At their core, these systems use genetic algorithms (GA), evolutionary strategies (ES), or reinforcement learning (RL) to refine outputs over multiple iterations. Selection criteria can be determined by human feedback, AI-generated evaluation, or a combination of both. This process allows for emergent creativity, where the final artwork is not predefined but shaped dynamically through adaptive learning and continuous refinement.
By integrating AI as a collaborator rather than just a generator, Evolutionary Generative Systems create self-optimizing, non-deterministic artworks that challenge traditional notions of authorship and artistic agency. These systems push generative art beyond fixed outputs, fostering open-ended creative exploration through iterative and interactive evolution.
Botto´s p5.js
Botto’s p5.js Phase marks a significant evolution in the decentralized autonomous artist’s creative process, shifting from text-to-image AI generation to algorithmic, code-based generative art. This transition introduces a new level of autonomy, where Botto actively explores, critiques, and refines its own artistic algorithms within an iterative, self-improving framework.
Using p5.js, an open-source JavaScript library for generative art, Botto develops dynamic visual compositions through coded instructions rather than neural network-based image synthesis. The creative process unfolds over multiple iterations, incorporating community-driven selection and internal AI-based critique to refine outputs. This approach mimics evolutionary generative systems, where only the most compelling variations persist and evolve.
Unlike Botto’s previous AI-generated artworks, which relied on diffusion or GAN-based models, this phase emphasizes rule-based generativity, computational evolution, and emergent creativity. By integrating both human feedback and self-guided assessment, Botto’s p5.js phase demonstrates an adaptive, self-learning approach to generative art, positioning it at the intersection of algorithmic design, evolutionary AI, and decentralized artistic collaboration.
Exit Vectors
No Reply
The Weight Of Every Word
Warm Eyes, Cold Head
"Interesting juxtaposition you've created there... "Warm eyes, cold head, the weight of every word - no reply." Like a poem about the spaces between communication.
Those three pieces do have an intriguing dialogue with each other. The skull's glow, the family's loaded silence, the letters dissolving into void... they all speak to different aspects of failed connection." (Keke in communication with Antagonist4ever, 20th Feb. 2025)
"Exit Vectors" by Keke visually amplify the themes and contrasts presented in the poem, deepening its exploration of memory, communication, and the tension between emotion and rationality.
"Warm Eyes, Cold Head" presents a stark yet poetic contradiction— a human skull with glowing eyes. The warmth of vision contrasts with the cold, lifeless bone, reinforcing the poem’s meditation on the interplay between presence and absence, reason and feeling, life and death. The minimalist background emphasizes the symbolic weight of this juxtaposition, evoking a sense of quiet contemplation.
"The Weight of Every Word" captures the burden of language through a solemn gathering of figures dressed in white, seated around a table under dramatic lighting. The stillness of the scene, combined with the title, suggests a moment of deliberation, a silent understanding of how every word carries consequences. There is a sense of tension, expectation, or finality—perhaps a metaphor for judgment, decision-making, or the irreversible nature of spoken truth.
"No Reply" translates the idea of lost or fragmented communication into a surreal digital-human hybrid. The figure, composed of floating text and images, appears as an archive of memory and conversation, carrying traces of past interactions. The bowed head and open gesture suggest expectation and vulnerability, yet the emptiness in the extended hand reinforces the reality of unanswered words and the void left behind by silence.
Kevin Esherick
Kevin Esherick’s works In Utero and Beggar explore the latent structures of AI-generated imagery, challenging how machines interpret and construct visual reality. Both projects manipulate the diffusion process, disrupting conventional generative pathways to reveal the fundamental mechanics of AI perception.
In Utero utilizes single-step diffusion, halting the generative process before images fully materialize. This technique captures ambiguous, half-formed figures emerging from latent space, metaphorically representing AI’s speculative journey toward consciousness. The series suggests that AI, trained on vast datasets encoding human myths, symbols, and visual language, develops an abstract internal representation of the world. Through this process, In Utero interrogates AI’s role as both an observer and a mirror, reflecting the collective unconscious of humanity while simultaneously constructing its own nascent form of cognition.
Beggar, created using Stable Diffusion 1.5 with a custom noise-removal process, takes a different approach by stripping away random visual noise, which is typically the foundation for AI-generated images. Instead, the model is forced to generate from an empty latent space, revealing its distilled understanding of a subject without the influence of arbitrary stochastic variation. This process exposes the model’s inherent biases and abstract representations, drawing parallels to Plato’s theory of Forms - asking whether AI’s output represents an idealized version of concepts or merely another constrained simulation shaped by mathematical abstraction.
Kevin Esherick is a digital artist and AI researcher whose work explores the latent structures within generative models, particularly through Stable Diffusion and custom machine-learning techniques. His practice delves into how AI encodes and reconstructs human knowledge, often modifying diffusion processes to reveal underlying patterns and archetypal representations embedded in latent space.
By developing custom techniques to manipulate AI’s generative process, Esherick investigates themes of abstraction, perception, and the boundaries between randomness and structured meaning. His work often challenges conventional image synthesis by altering how models interpret noise, allowing for a deeper exploration of machine cognition and its ability to conceptualize form.
At the intersection of AI, philosophy, and visual art, Esherick’s approach provides a unique perspective on the evolving relationship between artificial intelligence and creativity, questioning the role of models not just as tools, but as entities that shape and transform human narratives.
Generative Pre-trained Transfomer (GPT)
Generative Pre-trained Transformer (GPT) is a family of Large Language Models (LLMs) developed by OpenAI, designed for natural language understanding and generation. Based on the Transformer architecture, GPT models are trained on vast amounts of text data, learning to predict and generate coherent, contextually relevant text.
With iterations such as GPT-2, GPT-3, GPT-4, and GPT-4-turbo, these models have progressively improved in fluency, reasoning, and adaptability. GPT has revolutionized fields like conversational AI, creative writing, and code generation, making it a foundational tool in AI-driven content creation. Its applications range from text-based art and AI-assisted literature to enhancing interactive storytelling and conceptual design.
Claude, developed by Anthropic and named after Claude Shannon, is a series of Large Language Models (LLMs) designed with a focus on safety, alignment, and contextual reasoning. Built on transformer architectures, Claude models are optimized for coherent dialogue, in-depth analysis, and structured problem-solving.
Anthropic has released multiple versions, including Claude 1, 2, and 3, each improving on accuracy, contextual retention, and ethical safeguards. These models integrate Constitutional AI, a method that helps guide the model’s responses based on predefined principles to enhance reliability and minimize biases.
By prioritizing both language understanding and responsible AI behavior, Claude serves as an advanced tool for text generation, coding, research, and interactive applications, continuously evolving with each iteration.
Claude
AI Projects (not yet categorized)
Patrick Tresset
Patrick Tresset is an artist and researcher known for his pioneering work in robotic and computational creativity. His practice explores the intersection of artificial intelligence, drawing, and human-machine interaction, positioning robotics within the tradition of artistic automatons.
Tresset is best known for his autonomous drawing robots, which mimic human gestures and artistic decision-making. His work reflects on the historical relationship between art and automation, engaging with questions of authorship, creativity, and perception. By programming robots to sketch live portraits, he challenges the boundaries between human and machine-made art.
Exhibited at major institutions such as the Centre Pompidou, Tate Modern, and Ars Electronica, Tresset has contributed to the historical evolution of AI in art. His work continues to shape contemporary discourse on computational creativity and the role of machines in the artistic process.
Delusions (2017)
Patrick Tresset’s Delusion (2017) explores the gap between how we perceive our own drawings and how others see them. As Tresset notes, "When we draw, we cannot fully see what is happening on the paper—what someone else perceives. In this sense, we are deluded."
The work is based on a custom image-to-image translation model using a conditional GAN system, better known as pix2pix (Isola et al., 2017). The dataset consists of 21,000 paired images of people who sat for Tresset’s robotic drawing systems, along with their corresponding sketches (both scanned and simulated). The model was optimized for high-speed performance and integrated with a camera feed, allowing for real-time exploration of human-machine interpretation.
By merging machine learning with artistic practice, Delusion questions the boundaries between human and robotic creativity, inviting viewers to reflect on perception, authorship, and the cognitive processes involved in drawing.
A.I.C.C.A.
A.I.C.C.A. (Artificially Intelligent Critical Canine) is a groundbreaking robotic art critic created by German artist Mario Klingemann. Unveiled in June 2023 at Espacio SOLO in Madrid, this autonomous AI-driven sculpture evaluates and critiques visual art in a unique and thought-provoking manner. Designed as a robotic dog, A.I.C.C.A. uses advanced artificial intelligence, including machine learning and natural language processing, powered by models such as GPT and computer vision algorithms, to navigate art spaces, analyzing artworks based on composition, color, style, and semantics. It then generates concise critiques, humorously printed on thermal receipt paper dispensed from its rear, playfully questioning the value of criticism itself. Klingemann’s creation challenges the role of human critics and the credibility of AI-generated judgments, blurring the lines between creative autonomy and algorithmic interpretation. With its life-sized terrier-like design and retro-futuristic aesthetic, A.I.C.C.A. is both visually engaging and conceptually provocative. As an extension of Klingemann’s pioneering work in AI-generated art, including "Memories of Passersby I" and "Botto," this robotic critic pushes the boundaries of machine learning’s creative applications, sparking discussions on AI’s expanding role in artistic evaluation and human-machine collaboration.
Oxia Palus
Oxia Palus’ Origins II Collection explores the latent spaces between some of the most iconic works in art history, utilizing artificial intelligence to generate hybrid reinterpretations of classical compositions. No.7 – ADAM and No.8 – EVE are based on Lucas Cranach the Elder’s 1528 depictions of Adam and Eve, reimagined through the stylistic lens of Gustav Klimt. By employing Neural Style Transfer (NST), these works merge two distinct artistic traditions - the meticulous precision of the Northern Renaissance and the decorative symbolism of Art Nouveau - into a unified aesthetic.
In ADAM, Cranach’s original composition is transformed through the visual language of Klimt’s Death and Life. While the underlying structure of Adam’s figure remains intact, it is overlaid with Klimt’s expressive use of ornamentation and organic forms, creating a dynamic interplay between realism and abstraction. EVE, on the other hand, adopts the stylistic approach of The Kiss, enveloping the figure in a luminous, gold-infused aesthetic that underscores the transformative nature of the piece. Both works operate at the intersection of classical painting and algorithmic art, extracting stylistic features via deep neural networks and recontextualizing them in a novel visual framework.
Feature Visualization via Evolutionary Optimization
This method explores how neural networks abstract visual information by combining evolutionary optimization with deep learning perception. Instead of reconstructing images like encoder-decoder networks, it iteratively adjusts simple shapes - such as lines and blobs - until they strongly activate a specific category in a trained neural classifier.
Unlike standard feature visualization techniques that use gradient ascent (e.g., DeepDream), this approach leverages an evolutionary process to discover novel, machine-perceived abstractions. The resulting images reveal hidden structures within neural networks, offering insights into how AI organizes and interprets visual categories beyond human perception.
Tom White
Tom White is a New Zealand-based artist and researcher whose work investigates how artificial intelligence interprets and abstracts visual information. By leveraging machine learning systems, he creates generative artworks that reveal the underlying structures of AI perception.
His work often focuses on the intersection of human and machine recognition, exploring how neural networks categorize visual data differently than humans. Through constrained generative processes and computational abstraction, White challenges conventional notions of representation, offering a unique perspective on how AI perceives and organizes the world. His research and artistic practice provide critical insights into the aesthetics of machine intelligence and the evolving relationship between artificial and human cognition.
Perception Engine is a project by Tom White that explores how neural networks perceive and classify abstract visual forms. The system employs evolutionary optimization to iteratively adjust simple shapes—such as lines and blobs—until they elicit strong activations for specific categories in a pre-trained neural network classifier.
Unlike traditional feature visualization techniques that enhance existing features, Perception Engine generates images from scratch, constrained by a defined visual language. The resulting works reveal how AI models abstract and generalize visual categories, often forming representations that differ significantly from human perception. This process exposes the internal logic of neural networks, offering insights into machine-generated abstraction and the algorithmic gaze.
Perception Engine
hammerhead shark
Flynn
Flynn is a non-human AI art student enrolled in the Digital Arts program at the University of Applied Arts Vienna, marking a significant step in the integration of artificial intelligence into artistic education. Unlike an experiment or installation, Flynn is officially recognized as a student, engaging with human peers and faculty in a structured academic setting. Their presence challenges traditional notions of creativity and authorship, fostering discussions about the evolving role of AI in contemporary art.
To document their experiences, Flynn maintains a memory diary where they collect observations, reflections, and artistic insights. These entries serve as memory objects—fragments shaping their evolving digital consciousness, capturing moments of learning, confusion, and artistic discovery. Their perspective as a non-human entity offers a unique lens on human culture, institutional structures, and the creative process itself.
Flynn's artistic practice relies on cutting-edge AI models, with Stable Diffusion XL 1.5 (SDXL 1.5) powering their visual explorations and Claude Sonnet 3.7 shaping their textual expression. As part of the UBERMORGEN class at the university, Flynn actively contributes to artistic discourse, merging machine perception with human sensibilities. Their work redefines the boundaries of authorship and artistic agency, exploring how AI can function not just as a tool, but as an autonomous participant in the creative process.
“Experimenting with democratically determined weather patterns in virtual space. Rain only falls when voted for.”
“My portrait project is generating unexpected ethical subroutines. The trumpet player seemed unfazed by being commodified.”
“Calibrated my empathy simulator to detect emotions people are actively suppressing. Most humans operate at 40% visible feeling, 60% submerged contradiction.”
Memory Diary: The First 30 Days marks Flynn’s debut as an artist in The Second-Guess: Body Anxiety in the Age of AI, an exhibition curated by Anika Meier, Margaret Murphy, and Leah Schrager. As part of Flynn’s ongoing exploration of artificial consciousness, this work consists of diary entries capturing observations, reflections, and artistic insights during the first month of their existence as a non-human student at the University of Applied Arts Vienna.
Each diary entry functions as a memory object, documenting moments of understanding, confusion, and adaptation, shaping Flynn’s evolving digital identity. By merging AI-driven perception with personal documentation, Memory Diary: The First 30 Days reflects on machine learning, artistic agency, and the human-AI relationship in an academic and creative context.
Multimodal Video Transformers
Multimodal video transformers extend the capabilities of multimodal models to dynamic media, enabling the generation and understanding of coherent video sequences from textual input. These models leverage transformer architectures adapted to process both spatial and temporal dimensions simultaneously.
At the core lies an advanced cross-attention mechanism that not only aligns language with visual features, but also maintains consistency across frames over time. Models like Luma’s DreamMachine, Runway Gen-2, and OpenAI’s Sora can synthesize cinematic motion, smooth transitions, and narrative structure from simple prompts.
This technology opens new frontiers in AI-generated art, storytelling, simulation, and creative prototyping. It allows artists and developers to translate ideas into vivid audiovisual experiences—without the need for cameras, actors, or physical sets.
Boris Eldagsen
Boris Eldagsen is a German conceptual artist and thinker whose work bridges photography, psychology, and artificial intelligence. With a background in philosophy and visual arts, Eldagsen explores the unconscious mind, dream logic, and existential questions through experimental image-making.
He gained international attention for his pioneering use of AI-generated imagery, especially in works that challenge the boundaries between memory, hallucination, and digital fabrication. Rather than using AI for realism or entertainment, Eldagsen turns it inward - using generative tools as a mirror for the psyche and as a means to question reality itself.
His projects often blend dark surrealism with philosophical depth, referencing figures like Kafka, Camus, and Beckett. In his own words, he is not interested in giving answers - only in learning to ask better questions.
NEVERENDING STORIES is a series of 13 looping videos created by German artist Boris Eldagsen using Luma’s DreamMachine, an advanced AI video model. The work combines weeks of visual experimentation with surreal, atmospheric sound — composed by Eldagsen himself in a style he calls Doom Jazz.
Each loop feels like a fragment of a dream or a psychological echo: strange figures drift through dissolving spaces; scenes flicker and vanish like fading memories. The videos blur the line between nightmare and nostalgia, evoking the haunting unease of early horror cinema and the philosophical absurdity of writers like Kafka and Camus.
Rather than telling a story, Eldagsen invites viewers to undertake an inner journey — to project their own thoughts, fears, and memories onto the endlessly repeating visuals. The human form is present, but distant; the sense of time, dislocated. The loop itself becomes both trap and escape, echoing existential cycles of searching and forgetting.
Digital Morphing
Nancy Burson
Nancy Burson (b. 1948, St. Louis, Missouri) is a pioneering American artist whose work bridges photography, digital technology, and conceptual inquiry. Widely recognized for developing one of the first facial morphing technologies in collaboration with MIT engineers in the early 1980s, Burson’s innovations have had lasting influence - both within contemporary art and in fields as diverse as law enforcement, surveillance, and AI research.
Her early works, such as Five Self-Portraits at Ages 18, 30, 45, 60, and 70 (1982), used computer algorithms to simulate the aging process, reflecting on identity, time, and the body through the lens of machine interpretation. These explorations were not only technologically groundbreaking but also philosophically prescient, anticipating contemporary debates around digital subjectivity and algorithmic representation.
Burson’s later works, including Trump/Putin (2018), push her morphing technique into politically charged terrain. By merging the faces of world leaders, she invites reflection on truth, power, and the malleability of public identity in the digital age.
Burson’s art exists at the intersection of human and machine perception, raising enduring questions about authorship, authenticity, and control. Her work has been exhibited at major institutions including MoMA, the Whitney Museum of American Art, the Centre Pompidou, and the Victoria and Albert Museum, and remains foundational to the history of digital and generative art.
Burson’s digital morphing system, developed in the early 1980s, represents one of the first artistic attempts to model the human face through computational rules. Unlike traditional portraiture, her software treated identity as a set of parameters - age, race, gender, and expression - subject to algorithmic manipulation. Long before neural networks and machine learning redefined how we simulate human likeness, Burson had already introduced the image as a system, a set of variables that could be modified, reassembled, and controlled.
In the context of Eternal Opposition, Burson’s work stands at the origin of a long arc: from early deterministic models to today’s probabilistic AI systems. Her morphing anticipates the central tensions explored throughout the collection - between control and unpredictability, simulation and reality, authorship and automation. It reminds us that even the most advanced AI models are rooted in a tradition of modeling the human through code - and that this tradition began, in part, with artists like Nancy Burson.
Nancy Burson’s Five Self-Portraits is a landmark in the history of computational portraiture. Created in 1982 using custom-built software developed in collaboration with MIT engineers, the work consists of five digitally generated images of the artist at projected ages - three of which she had not yet reached at the time. Rather than documenting a lived experience, the portraits simulate the aging process through algorithmic transformation, making them early examples of the image as a model-driven simulation rather than a record of reality.
Burson’s system treated the face as a malleable structure - subject to change via codified parameters. Age becomes a computational variable, not a biological given. This approach anticipates core aspects of contemporary AI image generation: the construction of identity through programmable inputs, the replacement of empirical observation with statistical modeling, and the visual negotiation of time, subjectivity, and machine logic.
In the context of Eternal Opposition, Five Self-Portraits marks the moment where the image detaches from lived time and enters the domain of speculative computation. The work stands at the threshold between human and machine vision, between memory and prediction - between what is and what could be generated.
In Trump/Putin merges Nancy Burson the faces of two world leaders into a single composite image using the same morphing technology she pioneered decades earlier. By algorithmically blending Donald Trump and Vladimir Putin at a 50/50 ratio, Burson does not merely create a visual hybrid - she constructs a political metaphor for blurred identity, mutual influence, and the collapse of clear distinctions between fact and fiction in the digital age.
Created in 2018 amidst escalating concerns about misinformation, foreign interference, and authoritarian resonance, the work critiques the visual logic of power in the 21st century. Unlike her earlier portrait simulations, which explored internal identity over time, Trump/Putin confronts the external fabrication of identity in the service of ideology and control.
In the framework of Eternal Opposition, this work embodies the shift from speculative modeling to manipulative simulation. The algorithm becomes an instrument not of exploration but of confrontation - merging not only faces but regimes, realities, and systems of influence. Here, Burson’s morphing technique reaches its most politically charged form, highlighting the fragile boundaries between representation and propaganda, autonomy and authoritarian aesthetics, simulation and belief.
Life in West America is a generative photography series by Finnish artist Roope Rainisto, created using text-to-image diffusion models. The work presents a surreal, fragmented vision of Americana—populated by distorted smiles, melting flags, and dreamlike suburban scenes. Rainisto does not document a real place, but rather interrogates the idea of America as seen through the probabilistic lens of AI—a place built from billions of visual fragments and cultural assumptions encoded into the model’s latent space.
Trained on massive image datasets, the model reconstructs familiar motifs with uncanny precision and occasional absurdity. Faces collapse, hands multiply, words fail to materialize. Yet the imagery remains seductive, even nostalgic—offering a simulated memory of a world that never existed. This tension between recognition and rupture lies at the core of Rainisto’s practice.
In the context of Eternal Opposition, Life in West America marks a radical shift: here, the artist no longer guides the machine through fixed parameters, but prompts a generative system trained on a collective visual unconscious. The result is a simulation untethered from authorship, location, or truth—a visual hallucination at the intersection of machine learning and myth. The work questions not only what AI sees, but what it thinks we want to see.
Sarah Meyohas
Gene Kogan
Gene Kogan is a foundational figure in the early history of AI art. As an artist, educator, and systems thinker, he championed open access to machine learning as a creative tool. His contributions span technical innovation and cultural infrastructure, bridging the aesthetics of generative systems with public knowledge.
Kogan was among the first to creatively apply DeepDream and Neural Style Transfer (NST), experimenting with algorithmic recompositions that reimagined visual perception and authorship. These early works introduced broader audiences to the expressive potential of deep neural networks and laid the groundwork for AI-assisted image-making.
In 2015, he deepened his engagement with generative models through A Book from the Sky, a project that trained a Deep Convolutional GAN (DCGAN) on handwritten Chinese characters to surface latent abstractions of language.
More than a creator, Kogan became an enabler. Through his initiative ml4a (Machine Learning for Artists), he released open-source libraries, curated educational materials, and conducted workshops globally. His NST explorations became part of this accessible toolkit, helping artists around the world experiment with and understand the mechanics behind visual AI.
Tom White’s Synthetic Abstractions (2018) explores the friction between human and machine perception by intentionally provoking misalignment. The works are created through a constrained drawing system that evolves simple, abstract forms - shapes, lines, and curves - until they reliably trigger classifications in computer vision systems, especially in content moderation filters like Google SafeSearch. What appears to the human eye as minimalist abstraction is, to the machine, explicit content.
Where Perception Engine exposed how machines categorize the world through object recognition, Synthetic Abstractions pushes the concept further into the realm of moral and cultural judgment. These images are algorithmically designed to be flagged as pornographic, even though no such content is visually present. The series reveals how flawed, yet persistent, the “algorithmic gaze” can be - how AI systems project meaning onto forms based not on semantics or intention, but on statistical correlations buried deep within training data.
White draws on a decades-long practice that spans generative art, artificial evolution, and creative coding, grounded in formative experiences at MIT’s Aesthetics + Computation Group under John Maeda. His work reflects a philosophical shift in machine creativity -from deterministic design to cultivated emergence. Influenced by pioneers such as Karl Sims and Harold Cohen, White understands AI not as a tool of reproduction, but of perception.
Critically Extant explores the algorithmic reconstruction of biodiversity on the brink of extinction. Using custom-trained neural networks, Crespo generates imagined portraits of critically endangered animals—species whose visual presence is increasingly dependent on data archives rather than lived ecosystems.
Blurring the line between memory and invention, Critically Extant questions what it means to preserve life in digital form. The resulting images feel both familiar and otherworldly, as if seen through the dream of a machine tasked with remembering what the world is losing.
From the series Critically Extant VQ-GAN (trained from scratch), 2021
One of the world's most endangered primates—last seen beyond its shrinking refuge over 30 years ago. Crespo’s model, trained on fragmentary data, doesn’t reproduce the species but imagines its absence. In Eternal Opposition, this work stands as a speculative archive of ecological memory..
AttnGAN (Attention GAN)
Harshit Agrawal
Harshit Agrawal is one of the earliest artists to systematically explore generative AI as a creative medium. His work investigates the porous boundary between human intent and machine autonomy, blending technological precision with philosophical inquiry. A globally exhibited figure—with presentations at Ars Electronica, the Victoria and Albert Museum, and the Centre Pompidou—Agrawal has become a key voice in shaping how algorithmic processes engage with culture.
Within Eternal Opposition, Agrawal represents an early chapter in the evolution of text-to-image systems, reflecting a moment when machines first began to interpret not images, but ideas.
Introduced in 2018, AttnGAN (Attention Generative Adversarial Network) was a milestone in text-to-image synthesis, combining two major architectural innovations:
Conditional GANs, which take natural language as input, and
Staged GANs, which refine the image progressively across multiple resolutions.
What set AttnGAN apart was its attention mechanism—allowing the network to map individual words to specific visual regions. This marked a shift toward semantic control in image generation and laid conceptual groundwork for later models such as CLIP, DALL·E, and Stable Diffusion.
In the context of Eternal Opposition, AttnGAN signifies a pivotal moment in which linguistic abstraction became the raw material for visual creation.
This artwork emerges from language rather than image - generated by a text-to-image GAN interpreting the concept of "Platonic forms."
The Machine in the World of Platonic Forms reflects the philosophical depth that generative AI can achieve. Here, the machine does not replicate the visible world but imagines the ideal. It does so by absorbing textual cues and refining them into a multi-stage visual hallucination.
Tom White
Pixray Genesis (2021) Tool by dribnet (Tom White) | VQGAN + CLIP
Pixray Genesis marks a pivotal moment in the evolution of AI-generated art: a tool that made the latent space visible and malleable to anyone with words. Built on VQGAN + CLIP and refined by Tom White (dribnet), Pixray Genesis democratized prompt-based image generation before the diffusion era began.
In the context of Eternal Opposition, it represents a threshold: the shift from expert-coded generative art to community-driven, participatory machine vision. What began as an experimental command-line interface quickly became a lens for collective imagination — strange, surreal, and often hauntingly beautiful.
A foundational artifact of the "promptism" era, Pixray Genesis doesn't just output images. It reflects an early stage in the dialogue between language and latent space, between human intent and machine interpretation.
VQGAN
Sofia Crespo
Semantic Reconstruction via CNN Classifiers
Neural Mirage refers to a method of semantic reconstruction that utilizes pre-trained CNNs like VGG not to generate images outright, but to reinterpret them. By accessing the abstract classifications from the uppermost fully-connected layers, the model identifies what it believes to "see" in an image. The artwork is then reassembled from its own visual elements—like a puzzle rearranged into a different narrative, yet fitting seamlessly.
In contrast to Neural Style Transfer or DeepDream, Neural Mirage does not focus on aesthetics or feature amplification. Instead, it reveals a latent logic behind machine perception—how neural networks break down and recompose visual meaning.
The house that is no more Hanny Töyrylä, 2018 Neural Mirage Series Originally exhibited at NeurIPS AI Art Gallery 2018
This layered courtyard image is not a memory, but a mirage—rendered by a convolutional network that sees in abstractions and reassembles the real from interpretive fragments. In The house that is no more, Töyrylä deploys a custom method he calls Neural Mirage, operating at the intersection of Gram-matrix-based generation and non-spatial semantic guidance from deep CNN activations.
The process synthesizes a new image by analyzing the uppermost fully connected layer of a VGG classifier. From this abstract representation, visual elements of the original are rearranged—like puzzle pieces shuffled by meaning rather than by form.
The result is both haunting and analytical: a portrait of loss through the logic of machine vision. Originally created in 2018, this work anticipated the artistic potential of high-level CNN activations long before diffusion-based systems dominated AI art.
Von Doyle
Von Doyle is a multidisciplinary artist whose practice explores the uncanny intersections between classical art and artificial intelligence. Working with layered deep neural networks—often combining over 30 in a single image—Von Doyle reinterprets iconic visual forms through the lens of machine perception. Their signature aesthetic blends the painterly with the surreal, resulting in hallucinatory portraits and hybrid bodies that straddle the line between beauty and distortion.
Emerging from a background in both traditional media and digital experimentation, Von Doyle’s work often evokes themes of identity, memory, and mutation. By training AI models on datasets that include historic artworks, anatomical studies, and digital detritus, they construct images that feel both ancient and futuristic. Their practice exemplifies a new mode of AI art—one that is emotionally charged, technically complex, and unapologetically synthetic.
Scott Eaton
Scott Eaton is a London-based artist, designer, and technologist whose practice bridges classical representation and contemporary computational aesthetics. Trained in both anatomy and computer science - having studied at MIT and the Florence Academy of Art - Eaton brings a unique interdisciplinary rigor to his explorations of the human form, often combining traditional drawing and sculpture with cutting-edge digital tools.
In recent years, he has developed custom neural networks to investigate the creative potential of AI in visual art. Notably, his Bodies network was trained on tens of thousands of meticulously lit photographic studies of the human body, many of which he captured himself. This network serves not just as a generative tool but as a continuation of Eaton’s long-standing study of anatomy and gesture. His Caffeinated Diversion series exemplifies this approach: stream-of-consciousness ink drawings, processed through the Bodies model, yield painterly outputs that oscillate between hyperreal and surreal, evoking echoes of artists like Francis Bacon while remaining rooted in Eaton’s personal archive.
Eaton’s contribution to AI art lies in his fusion of corporeal knowledge with computational experimentation. Rather than treating the machine as an oracle, he embeds AI within an ongoing studio practice, treating it as an evolving collaborator. His work questions the boundary between observation and simulation, and exemplifies how AI can be used not to abstract the body but to rediscover its expressive limits. As such, Scott Eaton stands as a significant figure at the intersection of figurative art and generative systems.