How artists
work with AI.
An annotated source library, from a painted canvas to the model, archive and decisions behind it.
67 annotated sources · 66 opened · 1 search-only. Includes ten additional books and catalogues.
The earlier graphics survey was incomplete. This expansion prioritises accounts of how art is made, practical methods and substantial reading. Technical complexity is not evidence of artistic value.
Begin with the painter’s decisions.
These are suggested routes through the material. They are editorial selections, not evidence that a particular result will suit Iana.
Understand what to ask
Then investigate a workflow
Browse the annotated sources.
67 of 67 sources shown
Painting as a recursive feedback loop: Isle of Vitr∴ous
WangShui; interview by Sarah Cascone
Medium: oil on abraded aluminium. AI produces new images from previous paintings. WangShui sketches, collages and abrades those images into metal, adds thin oil layers, and improvises throughout; programmers collaborate on the systems.
Practical use, access and limits
Practical use: A generated image can be an intermediate drawing that an artist materially transforms.
Access: free interview text opened
Limits: The painting workflow is explicit; its exact model and training recipe are not. Process relevance does not establish aesthetic fit.
Tags: painting · oil · aluminium · drawing · collage · human finishing · recursive workflow
How I taught an AI to think like a painter
David Salle; with technologist Grant Davis
Medium: paintings over printed canvas backgrounds. Salle describes staged training-data choices and an image-variation control in Wand. He selects promising generated scenes, prints them onto canvas, and paints figurative layers over them.
Practical use, access and limits
Practical use: Separate the generated underlayer from the artist's decisions about edges, colour, composition and painted intervention.
Access: free studio-hosted PDF; six pages opened
Limits: His account includes other painters in early training, then his own work. It is not proof that five photographs suffice.
Tags: painting · canvas · printed underlayer · human finishing · artist archive · Wand
Art and Software Run as Parallel Tracks in the Life of Helena Sarin
Helena Sarin; AOTM interview
Medium: artist books and ceramics. Sarin trains models on her art and photography, extends pipelines through post-processing, and edits images into books. She describes potteryGAN as returning digital AI designs to physical artifacts.
Practical use, access and limits
Practical use: Treat generation, editing, sequencing and making an object as separate authored stages.
Access: free interview page opened
Limits: The interview explains the overall pipeline; it does not provide a reproducible clay construction, glaze or firing recipe.
Tags: ceramics · artist book · GAN · own photographs · post-processing · physical output
Human Allocation of Space and Figures: documented drawing-to-object processes
Scott Eaton
Medium: hand drawing, archival prints and bronze. Custom GANs translate Eaton's drawings into rendered bodies or geometric volumes. He authors source photography and detailed drawings; sculpture output is converted to polygons, printed, moulded and cast.
Practical use, access and limits
Practical use: A clear hand-drawn structure can remain central while a learned system supplies a different representational layer.
Access: free artist presentation opened
Limits: Figurative/anatomical methods do not establish abstract-painting fidelity; substantial custom datasets and fabrication are involved.
Tags: drawing · photography · bronze · print · image-to-image · human structure
Shared Visual Abstractions
Tom White
Medium: abstract ink prints. A constrained drawing system searches stroke placement, thickness and colour using responses from trained image classifiers. White designs the drawing vocabulary and experiments; photographs of physical prints test recognition across models.
Practical use, access and limits
Practical use: Restricting marks and colours can be an artistic decision, independent of a generator's realism.
Access: open paper; HTML methods opened
Limits: The paper measures classifier responses and a small human label-matching survey; neither measures beauty or usefulness to Iana.
Tags: abstraction · ink print · constraints · classifier feedback · research paper
In Silico / Technical Images
Casey Reas; Erika Weitz on Technical Images
Medium: wet-plate collodion photographs, aluminium prints and software. Reas curates plant scans, microscopy and landscape imagery for GAN training and selects outputs. With Weitz, generated images become collodion plates; RGB channels are separately photographed and recombined.
Practical use, access and limits
Practical use: Physical translation and channel-level recomposition can contribute as much as image synthesis.
Access: free artist process page opened
Limits: Botanical imagery is contextual here, not an aesthetic recommendation. The page documents stages without a complete executable workflow.
Tags: photography · collodion · aluminium · physical output · dataset curation · GAN
Fall of the House of Usher I and II
Anna Ridler
Medium: ink drawing dataset and animation. Ridler makes more than 200 drawings after the 1929 film; interconnected GANs generate stills and feedback material for subsequent training sets. The handmade dataset is also an exhibited work.
Practical use, access and limits
Practical use: Preparatory drawings and dataset construction can remain visible parts of the finished project.
Access: free museum page; artist-authored text opened
Limits: This is intentionally horror/uncanny work, included for process rather than its aesthetic; implementation details are limited.
Tags: drawing · ink · animation · handmade dataset · archive · horror aesthetic
My Artificial Muse
Albert Barqué-Duran, Mario Klingemann and Marc Marzenit
Medium: live oil painting, projected neural generation and sound. A neural network produces a classical-style image; Barqué-Duran interprets it in paint. Embodied sensors connect his movement with a generative soundscape.
Practical use, access and limits
Practical use: Painting can be a selective human interpretation of an AI proposal, rather than a literal transfer.
Access: free artist project page opened
Limits: The current project page inconsistently says 2016 and 2017; a collaborator account and the artist exhibition list document a 2017 performance. The inception/premiere conflict remains unresolved. Model settings and a painting recipe are absent.
Tags: oil painting · performance · human interpretation · neural generation · sound
Reflection: cameras and physical feedback in robotic painting
Pindar Van Arman; artist-hosted essay by centoshi
Medium: paintings on stretched canvas. Cameras observe marks and multiple algorithms evaluate the changing image before the robot makes further marks. Physical paint variation becomes input; Van Arman authors and revises the robotic system.
Practical use, access and limits
Practical use: The material surface can feed back into computational decisions instead of merely receiving a finished digital image.
Access: free artist-hosted process essay opened
Limits: Claims about machine creativity are the artists' interpretation. The page does not supply a reproducible control implementation.
Tags: painting · robotics · camera feedback · physical material · canvas
Tabula Rasa: an artist's account of training on solid colours
David Young
Medium: abstract generated images. Young begins GAN training with one black image, then adds further solid colours and observes resulting textures, grids and shapes. He authors the restricted material and compares stages of the system's development.
Practical use, access and limits
Practical use: An AI investigation can begin with colour and compositional constraints, without imitating a painter's complete style.
Access: free artist essay opened
Limits: Machine emotion is an acknowledged metaphor, not a demonstrated mental state; the essay is not a current software tutorial.
Tags: abstraction · colour · minimal dataset · GAN · artist essay
Hallucinations / Adversarially Evolved Hallucinations
Trevor Paglen
Medium: synthetic images and dye-sublimation prints. Paglen constructs training categories from literature, history and other conceptual corpora, trains recognition systems, then uses adversarial generation to produce images satisfying those systems.
Practical use, access and limits
Practical use: Choosing what categories a system can recognise can be a substantive part of the artwork's argument.
Access: free artist page opened
Limits: The artist explains the conceptual pipeline; architecture and training details are absent. Some imagery is disturbing.
Tags: print · dataset authorship · classification · GAN · critical practice
Zizi — Queering the Dataset
Jake Elwes
Medium: multichannel video. Elwes retrains a StyleGAN model originally trained on FFHQ with 1,000 additional images of drag and gender-fluid faces. Dataset intervention changes the model's visual vocabulary; Elwes authors framing and installation.
Practical use, access and limits
Practical use: Dataset additions can deliberately challenge what a pretrained model treats as normal.
Access: free artist project page opened
Limits: Historical online-image collection is described, not offered as a permissions precedent. No exact retraining recipe is provided.
Tags: video · portrait · dataset intervention · drag · StyleGAN · critical practice
Wearing, Gillian: museum production account
Gillian Wearing with Wieden + Kennedy
Medium: five-minute colour film. Wearing casts more than fifteen performers to portray her; their faces, bodies and voices are digitally combined with hers into a deepfake persona. Casting, scripting and performance remain human-authored.
Practical use, access and limits
Practical use: AI-mediated identity can be built around authored performances and a script, rather than anonymous generated footage.
Access: free museum PDF; relevant work entry examined
Limits: The relevant wall-text entry was read, not the entire 82-page PDF. The neural model and compositing pipeline are unspecified.
Tags: film · performance · deepfake · identity · museum documentation
Dio: Media and Materiality
Ben Snell; Phillips interview/account
Medium: physical sculpture. Snell trains a system with 3D models of classical sculpture, then makes its generated form from the powdered components of the computer itself. He authors training selection, fabrication and the material narrative.
Practical use, access and limits
Practical use: The output material can express the concept behind an AI process rather than simply illustrate its result.
Access: free interview/account opened
Limits: The source describes classical 3D training models and casting computer powder mixed with clear resin; architecture and a complete fabrication specification are absent. Its autonomy language is promotional.
Tags: sculpture · 3D dataset · material concept · physical output
Cloud of Petals: artist conversation on collection, film and VR
Sarah Meyohas; interview by Maisie Skidmore
Medium: photography, 16mm film, a pressed-petal archive and VR. Workers photograph 100,000 petals and select favourites; Meyohas films their labour and works with ML specialists on generated petals. Human selection persists in the physical archive.
Practical use, access and limits
Practical use: Source collection, labour and subjective selection can become visible elements of the artwork.
Access: free artist conversation opened
Limits: This does not demonstrate botanical imagery suits Iana. The source documents stages but omits training parameters.
Tags: photography · archive · film · VR · data collection · human selection
Not the Only One (N'TOO)
Stephanie Dinkins
Medium: voice-interactive sculpture/archive. Oral histories from three generations of Dinkins's family and culturally attuned material inform a conversational AI. The artist records, selects and aligns material, designs the entity, and iterates its model and sculptural embodiment.
Practical use, access and limits
Practical use: An authored archive may include voices and histories, with usefulness judged against the represented people's needs.
Access: free artist project and technical notes opened
Limits: Dinkins describes imperfect responses and model changes. This is an archival/social practice, not evidence for painting-style transfer.
Tags: oral history · archive · sculpture · NLP · community · small data
The Call
Holly Herndon and Mat Dryhurst; Serpentine Arts Technologies and participating choirs
Medium: choral sound installation. A composed songbook and recording protocol guide sessions with fifteen community choirs; recordings train choral AI models. Artists, singers, researchers and designers author different stages, with contributor governance explored through a Data Trust.
Practical use, access and limits
Practical use: Design source-making and contributors' roles before treating a model as the artwork's sole author.
Access: free institutional project page and credits opened
Limits: Sound and collective governance are the transferable ideas; this is not a visual painting workflow or a checked public model service.
Tags: sound · participation · authored dataset · consent · choral AI · installation
Artificial Life: a process interview
Sofia Crespo; interview by Andrew Dubber, MTF Labs
Medium: synthetic-organism imagery and video. Crespo explains collecting subject-specific datasets, training separate caterpillar and butterfly models, then connecting their nearest visual states. She authors datasets, transitions and selection among many generated alternatives.
Practical use, access and limits
Practical use: Separate models or collections can provide different visual vocabularies whose relationship the artist composes.
Access: free podcast transcript opened; audio not auditioned
Limits: Organic/uncanny forms are contextual, not recommended aesthetics. The transcript was examined; no model or audio was run.
Tags: video · dataset curation · synthetic biology · GAN · organic aesthetic
The Substitute
Alexandra Daisy Ginsberg; animation by The Mill
Medium: paired video installation. A rendered rhino's behaviour is based on DeepMind agent research; animal behaviours and sounds are adapted from research footage. Ginsberg and collaborators author the ecological premise, animation and dual-screen presentation.
Practical use, access and limits
Practical use: AI-derived behaviour can serve a specific conceptual question rather than operate as decorative movement.
Access: free artist project page opened
Limits: The artist credits animation based on research; this page does not establish a live-learning system in the gallery.
Tags: video · animation · agent research · ecology · contextual reference
Feminist Data Set: toolkit and process
Caroline Sinders
Medium: participatory workshops, archive and toolkit. People define categories and submit material; Sinders reviews submissions and adjusts subsequent collection to address exclusions. The artwork examines human decisions throughout a future machine-learning pipeline.
Practical use, access and limits
Practical use: Collecting, naming, rejecting and documenting source material can be artistic work in its own right.
Access: free artist toolkit PDF; process and workshop sections examined
Limits: The reader explicitly presents a complete feminist AI as prospective. This is upstream AI-art practice, not a finished generative-image system.
Tags: archive · workshop · data curation · participation · upstream practice · critical practice
Digital Art, fourth edition
Christiane Paul
A historical and critical survey covering digital media's relationship to painting, print, photography and sculpture, alongside AI, installation and collecting.
Practical use, access and limits
Practical use: Use it to choose an artistic lineage and presentation medium before choosing software.
Access: Publisher metadata and description opened; sample-spread links present but not inspected; full book not retrieved.
Limits: Broad digital-art history, with AI as one topic; not a contemporary image-generator or painting manual. Assessment is based on the publisher description.
Tags: AI and adjacent digital art · history · theory · painting · physical output
Artificial Aesthetics: Generative AI, Art and Visual Media
Lev Manovich and Emanuele Arielli
Essays examine creative agency, aesthetic judgment, perception, media history and the status of AI images.
Practical use, access and limits
Practical use: Create a studio rubric separating visual appeal, personal intention, authorship and useful surprise.
Access: Free author-hosted 194-page complete PDF retrieved; front matter, contents and selected passages inspected; not read cover-to-cover.
Limits: Critical theory rather than a software manual. The author requests a 2024 citation while dating the collected PDF January 15, 2025. Not read cover-to-cover.
Tags: AI art · open access · theory · aesthetics · evaluation
The Perception Machine: Our Photographic Future between the Eye and AI
Joanna Zylinska
Connects computational photography and machine vision with perception and the author's photographic, collage, film and installation projects.
Practical use, access and limits
Practical use: Examine how photographic references and machine interpretation change a painting's point of view.
Access: Publisher metadata and OA notice opened; full-text endpoint returned 403 on 3 November 2026; no full book retrieved.
Limits: Theory and artistic inquiry, not a painting workflow. Open-access status is confirmed by MIT Press; the full book was not retrieved or read.
Tags: AI art · machine vision · critical theory · photography · collage
Computational Formalism: Art History and Machine Learning
Amanda Wasielewski
Studies how artwork datasets, style categories and machine vision shape computational art-historical analysis.
Practical use, access and limits
Practical use: Question whether a model's apparent stylistic match captures the artist's decisions or merely recurring visual features.
Access: Publisher metadata and OA notice opened; DiVA PDF routes timed out and institutional record denied access on 3 November 2026; no book full text retrieved.
Limits: AI analysis of art is the central subject, rather than making paintings. Publisher description checked; full book not read.
Tags: AI and art analysis · datasets · style · critical theory · evaluation
Artificial Intelligence and the Arts: Computational Creativity, Artistic Behavior, and Tools for Creatives
Penousal Machado, Juan Romero and Gary Greenfield, editors
Technical collection spanning visual art, music, design and games; visual-art chapters include artificial life, search-based art and neural style transfer.
Practical use, access and limits
Practical use: Ask a technical collaborator which model family fits a specific studio task instead of treating all generative systems alike.
Access: Publisher overview and table of contents opened; full chapters/book not retrieved.
Limits: Research-oriented and partly mathematical; contents were checked, not complete chapters. Predates today's image-generator interfaces.
Tags: AI art · technical reference · style transfer · evolutionary art · research
Computers and Creativity
Jon McCormack and Mark d'Inverno, editors
Interdisciplinary collection joins computational creativity, aesthetic evaluation and artists' approaches across visual, musical and interactive work.
Practical use, access and limits
Practical use: Compare designing a system, selecting its outputs and making the final material work as different creative decisions.
Access: Publisher overview and table of contents opened; full chapters/book not retrieved.
Limits: Foundational computational-art and creativity research, including AI and evolutionary systems; not a diffusion-model guide. No full-book reading.
Tags: AI and adjacent computational art · creativity · evaluation · history · research
The Practice of Art and AI: European ARTificial Intelligence Lab
Andreas J. Hirsch, Markus Jandl and Gerfried Stocker, editors
Documents art-and-AI projects through the European ARTificial Intelligence Lab and Ars Electronica's art/science context.
Practical use, access and limits
Practical use: Find comparable projects and collaborators, then trace an individual project's process rather than copying its appearance.
Access: Publisher record opened; print marked permanently out of stock. Ars Electronica advertises a free 6 MB PDF in search extraction, but the download page and PDF returned 403 on 3 November 2026; full book not retrieved.
Limits: Publisher record verified; catalogue contents and full book were not inspected. Do not equate an advertised archive download with successful full-text retrieval.
Tags: AI art · artist projects · institutional practice · art and science · catalogue context
Digital Art: 1960s to Now
Pita Arreola, Corinna Gardner and Melanie Lenz
Decade-based history combines essays with interviews involving artists, curators, gallerists and collectors, from computer works on paper to contemporary digital practice.
Practical use, access and limits
Practical use: Study how artists move between drawings, systems, installations and institutions while retaining a coherent practice.
Access: Museum shop metadata and description plus V&A research page opened; sample-spread links present but not inspected; full book not retrieved.
Limits: AI is part of a broader history. The V&A research-project page confirms the 2024 publication; the book was not read in full.
Tags: AI and adjacent digital art · history · artist interviews · works on paper · presentation
Artistes & robots
Laurence Bertrand Dorléac and Jérôme Neutres, editors; Grand Palais Rmn
Exhibition catalogue provides a route into robot-assisted and autonomous art, complemented by the institution's educational materials.
Practical use, access and limits
Practical use: Compare the role of physical gestures, autonomy and audience experience before considering robotic output.
Access: Publisher catalogue record opened; associated 31-page teacher dossier and 8-page English visitor guide retrieved and sampled; catalogue itself not retrieved.
Limits: French-language catalogue; book record rather than full text inspected. The exhibition includes rule-based and robotic work beyond machine learning.
Tags: AI and adjacent robotic art · catalogue · physical output · history · French
Generative Art: A practical guide using Processing
Matt Pearson
Instructional introduction to algorithmic image making with Processing, including randomness, noise, emergence, autonomy and fractals.
Practical use, access and limits
Practical use: Build a small rule-driven study of line, spacing or rhythm as a comparison with an AI-assisted study.
Access: Publisher metadata and 14-page sample/contents opened; full book not retrieved; code not run.
Limits: Adjacent generative coding, not machine learning. Detailed steps are described by the publisher and contents; old code was not run or validated.
Tags: adjacent computational art · practical manual · creative coding · composition · Processing
Creative AI Lab: tools, resources and R&D Labyrinths
Serpentine Galleries and King's College London; R&D Labyrinths with Trust
Research programme links tool resources, discussions and artist-led software walkthroughs. R&D Labyrinths describes step-by-step poetry, composition and governance examples.
Practical use, access and limits
Practical use: Use its walkthrough format to document source material, interface choices and artistic decisions in a painting experiment.
Access: Free Serpentine Lab and R&D Labyrinths descriptions and KCL project page opened; videos, database entries and tools not tested. Programme page remains labelled ongoing.
Limits: The three observed walkthrough descriptions concern poetry, music and collective governance, not painting recipes. Historical tool links may need fresh compatibility checks.
Tags: AI art · process documentation · tools · artist walkthroughs · cross-media
Future Art Ecosystems 4: Art x Public AI
Serpentine Arts Technologies, drawing on Creative AI Lab research
Institutional briefing examines the AI stack and cultural organisations' ability to shape AI infrastructure.
Practical use, access and limits
Practical use: Before commissioning a studio system, identify who controls data, model access, compute and continuing support.
Access: Volume-specific Serpentine launch page and series overview retrieved; launch dated 19 March 2024. Full briefing reader could not be retrieved on 3 November 2026; briefing not read.
Limits: Infrastructure and cultural policy, not visual technique. Programme and launch date checked; full report was not retrieved or read. This is the 2024 FAE4 volume, not the latest series volume; the series page now also lists FAE5.
Tags: AI art · infrastructure · institutional practice · data · strategy
European ARTificial Intelligence Lab archive
Ars Electronica and European cultural/scientific partners
Historical programme archive indexes residencies, activities, partners and juries around artistic engagement with AI.
Practical use, access and limits
Practical use: Follow a residency from artistic question to scientific partnership and exhibition, looking for viable collaboration models.
Access: Free archive introduction opened; dynamic results remained Loading data in text extraction; no individual residency record or current application offer verified.
Limits: Completed programme, not a current residency offer. The landing page does not establish process detail for every indexed project or reusable code.
Tags: AI art · residencies · archive · art and science · collaboration
Renaissance 3.0: A Base Camp for New Alliances of Art and Science in the 21st Century
ZKM | Center for Art and Media Karlsruhe; curators Peter Weibel and Anett Holzheid
Exhibition archive follows artistic laboratories and shared art/science tools, with individual works and linked interviews.
Practical use, access and limits
Practical use: Consider how a technical experiment becomes a material artwork with a subject, duration and installation context.
Access: Free historical exhibition page, work catalogue index and the native picture object description opened; videos/talks not played; no current workshop availability established.
Limits: Broad art/science and computational practice, not exclusively AI. Historical workshops listed on the page are past events; no current attendance is implied.
Tags: adjacent computational art · art and science · archive · materiality · interviews
AI: More than Human
Barbican; curators Suzanne Livingston and Maholo Uchida
Exhibition places artistic commissions alongside scientific AI research and questions about human-machine relationships.
Practical use, access and limits
Practical use: Use artist interviews to compare the intended experience of an artwork with the technical system producing it.
Access: Free touring overview and original 2019 event archive opened; interview/essay links observed but media not played. Hire availability from mid-April 2027 is institutional touring information.
Limits: A survey rather than a production manual. The original 2019 exhibition is past; the current touring-hire page is not a current public exhibition listing.
Tags: AI art · exhibition · artist interviews · critical context · audience experience
Chance and Control: Art in the Age of Computers
Victoria and Albert Museum
Displays computer-generated prints, drawings and paintings by artists including Vera Molnár, Frieder Nake, Manfred Mohr and Roman Verostko.
Practical use, access and limits
Practical use: Compare controlled variation in line and composition with a hand-painted series, using actual works on paper as references.
Access: Free past-display overview and artwork highlights opened; 2-page conference programme retrieved (26 October 2018); no procedural manual or current enrollment.
Limits: Predominantly adjacent algorithmic art, not evidence about modern learned image models. Past display and conference; no procedural reproduction instructions.
Tags: adjacent computational art · history · works on paper · composition · archive
A Shift in Artistic Practices through Artificial Intelligence
Kıvanç Tatar, Petter Ericson, Kelsey Cotton, Paola Torres Núñez del Prado, Roser Batlle-Roca, Beatriz Cabrero-Daniel, Sara Ljungblad, Georgios Diapoulis and Jabbar Hussain
Leonardo article surveys how accessible AI systems challenge artistic roles, conventions and practices across media.
Practical use, access and limits
Practical use: Name which studio roles a system changes: proposing, selecting, composing, producing or performing.
Access: Author page and arXiv metadata opened; free 13-page author-deposited manuscript PDF retrieved and sampled. Former five-page UPF published-PDF endpoint returned 502 on 3 November 2026; publisher DOI returned 403.
Limits: Short field overview, not a controlled study demonstrating improved paintings or a complete production recipe. Published article and preprint dates differ.
Tags: AI art · journal · artistic practice · cross-media · open access
Creative Applications of Deep Learning with TensorFlow
Parag Mital; Kadenze
Syllabus describes dataset preparation, network building, generative models and creative applications using an early TensorFlow stack.
Practical use, access and limits
Practical use: Use its dataset-to-model-to-experiment structure when discussing a learning plan with a technical collaborator.
Access: Historical Kadenze syllabus and Enrollment Closed label verified only through search extraction; direct page fetch timed out on 3 November 2026. Author-owned archived CADL README and notebook index opened; video access/enrollment and code compatibility untested.
Limits: Historical technical course, not a verified current class. Canonical page lists TensorFlow 0.11.0; the separate -i offering lists 0.9.0. Dataset, network, DeepDream/style-transfer and generative-model lessons are historical instruction, not physical painting technique.
Tags: AI art · archived course · technical education · datasets · historical software
CODE 2024: Deep Nudes and Digital Oblivion
IMPAKT Centre for Media Culture and collaborating artists/partner organisations
Documents a collaborative art programme addressing AI images, digital identity, opaque models and control over personal data.
Practical use, access and limits
Practical use: Choose a concrete artistic question about representation or agency before building an AI intervention.
Access: Free exhibition archive and project list opened; individual project workflows not retrieved; physical exhibition and artist talks took place in 2024.
Limits: Critical media-art projects rather than painter-facing technique. A six-month co-creation process is documented, but the overview does not provide complete technical workflows.
Tags: AI art · critical practice · digital identity · collaboration · archive
InvokeAI: Crafting images with the Canvas
InvokeAI contributors
A raster painting, mask and regional guidance become candidate edits, staged for acceptance into layers.
Practical use, access and limits
Practical use: The artist draws masks, adjusts denoising and accepts or discards results. An abstract painter could isolate one unresolved passage.
Access: free documentation; application/model setup not tested
Limits: Versioned v7 documentation; controls depend on model support. Documented masking does not establish fidelity to an individual practice.
Tags: artist-directed-editing · own-reference-style · hands-on · layers · inpainting
ComfyUI inpainting and outpainting examples
ComfyUI project
An image and alpha mask feed an editable node workflow; padding extends its bounds for outpainting.
Practical use, access and limits
Practical use: The artist chooses the editable region and inspects workflow parameters. Useful for making variations around a retained painted fragment.
Access: free examples with embedded workflows; checkpoints not downloaded
Limits: These examples name older v2/AnythingV3 checkpoints. Loading workflow metadata is documented; compatibility with a fresh installation was not tested.
Tags: artist-directed-editing · hands-on · inpainting · outpainting · embedded-workflow
Photoshop: Use reference images for consistent results
Adobe
A selection plus uploaded reference guides Generative Fill; the guide distinguishes whole-image/object references and placing/swapping content.
Practical use, access and limits
Practical use: The artist supplies both region and example. Try one approved motif as reference, then compare candidates against a manually composed study.
Access: free instructions; Photoshop and generation access separate
Limits: Commercial application and model-specific options; this manual demonstrates controls, not exact copying, style fidelity or a verified free generation entitlement.
Tags: artist-directed-editing · own-reference-style · hands-on · reference-image · commercial-tool
Draw Things Basic UI guide
Draw Things community documentation contributors
The guide covers canvas drawing, imported images, inpainting, seeds, project history and configuration templates.
Practical use, access and limits
Practical use: The artist retains drawing, erasing, strength and seed choices. It offers a route from a rough colour sketch to saved alternatives.
Access: free community guide hosted by tool creator
Limits: Community-authored, with an explicitly old UI section. Treat historical resolutions and button locations as version-specific; application and models were not tested.
Tags: artist-directed-editing · hands-on · sketch-to-image · local-workflow · community-authored
Diffusers: Image-to-image
Hugging Face Diffusers maintainers
A starting image and prompt produce a transformed image; examples vary strength, guidance and model family.
Practical use, access and limits
Practical use: The artist chooses the initial block-in and change amount. A small strength sweep could reveal when useful variation becomes unwanted rewriting.
Access: free documentation with executable code examples; not executed
Limits: Python/model setup is required. Examples use specific diffusion architectures; a strength value is not a portable percentage of artistic fidelity.
Tags: artist-directed-editing · hands-on · image-to-image · parameter-study
SDEdit: Guided Image Synthesis and Editing with Stochastic Differential Equations
Chenlin Meng and colleagues
Coloured strokes or a rough composite are noised and denoised into image variants using a pretrained diffusion prior.
Practical use, access and limits
Practical use: The artist supplies the arrangement and edits; its noise-versus-retention tradeoff explains why a block-in can guide without precisely constraining output.
Access: open paper, project and original code; not executed
Limits: Original checkpoints target faces, bedrooms and churches. Realism-oriented results do not establish suitability for preserving abstract marks.
Tags: artist-directed-editing · sketch-to-image · foundational · image-to-image
IP-Adapter: Text Compatible Image Prompt Adapter for Text-to-Image Diffusion Models
Hu Ye and colleagues; Tencent AI Lab
A reference image conditions generation alongside text, without training a new adapter for each supplied image.
Practical use, access and limits
Practical use: The artist chooses references and image influence. Compare reference-guided motif or palette studies before committing to training on an archive.
Access: open paper and official code/notebooks; not executed
Limits: Adapters must match their base architecture. Original defaults center-crop references, potentially losing peripheral marks; similarity is not an authorship/style guarantee.
Tags: own-reference-style · artist-directed-editing · reference-image · adapter
DreamBooth: Fine Tuning Text-to-Image Diffusion Models for Subject-Driven Generation
Nataniel Ruiz and colleagues; Google Research
A selected subject-image set fine-tunes a model, associating an identifier with that subject for new renditions.
Practical use, access and limits
Practical use: The artist authors the training set and judges retained identity. Useful background for recurring personal motifs, rather than automatic style certification.
Access: open original paper/project; separate Diffusers training implementation opened
Limits: Original work used Imagen; linked Diffusers training is a separate implementation. Few-image subject experiments do not establish a universal painting-dataset minimum.
Tags: own-reference-style · personalization · training · foundational
Training Generative Adversarial Networks with Limited Data / StyleGAN2-ADA
Tero Karras and colleagues; NVIDIA
A curated dataset trains or fine-tunes a GAN; adaptive augmentation addresses discriminator overfitting under limited data.
Practical use, access and limits
Practical use: The artist controls archive selection and sample curation. Useful for understanding dataset-shaped visual worlds and interpolation, beyond text prompting.
Access: open paper and official training code; not executed
Limits: Historical CUDA/PyTorch stack. Limited-data experiments are not evidence that five artwork photographs reliably capture an artist’s practice.
Tags: own-reference-style · training · gan · dataset-authorship
InstructPix2Pix: Learning to Follow Image Editing Instructions
Tim Brooks, Aleksander Holynski and Alexei A. Efros
An existing image plus an editing instruction produces an edited image without per-image fine-tuning or inversion.
Practical use, access and limits
Practical use: The artist specifies a change and rejects collateral changes. Compare a single palette/material instruction with a manually painted counterpart.
Access: open paper, project and official code; not executed
Limits: The project shows excessive edits and failures to isolate objects or reorganize scenes. Its 2023 model is a historical baseline.
Tags: artist-directed-editing · instruction-editing · foundational
DragDiffusion: Harnessing Diffusion Models for Interactive Point-based Image Editing
Yujun Shi and colleagues
An image, editable mask and handle/target points produce a spatially deformed image through latent optimization.
Practical use, access and limits
Practical use: The artist chooses where a form moves. Potentially useful for studying spacing or silhouette shifts without describing them in prose.
Access: open paper, project and official UI code; not executed
Limits: Original real-image workflow first trains an image-specific LoRA; documented setup targets NVIDIA/Linux and an older Diffusers version.
Tags: artist-directed-editing · hands-on-research · point-editing · composition
Plug-and-Play Diffusion Features for Text-Driven Image-to-Image Translation
Narek Tumanyan, Michal Geyer, Shai Bagon and Tali Dekel
A source image and target prompt yield a translated image, with source features injected to retain semantic layout.
Practical use, access and limits
Practical use: The artist supplies composition and chooses structural influence. Compare surface or colour alternatives while checking that key relationships survive.
Access: open paper, project and official code; not executed
Limits: Original code targets Stable Diffusion v1.4. Layout preservation can leak source appearance; it is not exact contour or brushwork retention.
Tags: artist-directed-editing · structure-control · image-to-image
Paint by Example: Exemplar-based Image Editing with Diffusion Models
Binxin Yang and colleagues
A source image, region mask and exemplar image become a reference-conditioned inpainting result.
Practical use, access and limits
Practical use: The artist supplies where and what visually. An approved motif or material study can communicate changes that are difficult to describe.
Access: open paper and official code with inference examples; not executed
Limits: The model intentionally transforms exemplars rather than pasting pixels. Original checkpoints and environment require inspection; exact reference fidelity is unproven.
Tags: own-reference-style · artist-directed-editing · reference-image · inpainting
SemanticDraw: Towards Real-Time Interactive Content Creation from Image Diffusion Models
Jaerin Lee and colleagues
Hand-drawn regions receive separate semantic prompts, producing an evolving raster image through a streaming diffusion pipeline.
Practical use, access and limits
Practical use: The artist paints regions and changes their meaning. Promising for testing balances of differently described areas in a composition.
Access: open paper and official code with demos; not executed
Limits: Speed claims depend on published hardware/settings. Model-specific SD1.5, SDXL and SD3 demos are documented; arbitrary-model compatibility was not tested.
Tags: artist-directed-editing · hands-on-research · semantic-brushes · regional-control
PromptPaint: Steering Text-to-Image Generation Through Paint Medium-like Interactions
John Joon Young Chung and Eytan Adar
Prompt mixtures, stencils and interventions steer different areas and stages of image generation through a paint-inspired interface.
Practical use, access and limits
Practical use: The artist mixes concepts and intervenes during generation. Valuable design precedent for gradual exploration rather than a single prompt submission.
Access: open author paper and code; publisher fetch returned 403
Limits: Research setup spans local services, Colab and ngrok, possibly paid compute. Interface studies do not establish superiority for abstract painters.
Tags: artist-directed-editing · co-drawing · interaction-design · prompt-mixing
Collaborative Neural Painting
Nicola Dall’Asen and colleagues
User brushstrokes or an object class condition proposed stroke sequences, allowing iterative editing and composition.
Practical use, access and limits
Practical use: The artist can modify individual strokes during completion. Its editable representation is a useful precedent for preserving decisions beyond final raster pixels.
Access: open paper/project and official code; dataset release incomplete
Limits: Dataset remains marked Coming soon; checkpoint download is linked but untested. Object-painting research does not demonstrate an arbitrary-archive collaborator. The project page also marks the demo Coming Soon.
Tags: co-drawing · editable-strokes · artist-directed-editing · research-prototype
FRIDA: A Collaborative Robot Painter with a Differentiable, Real2Sim2Real Planning Environment
Peter Schaldenbrand, James McCann and Jean Oh
Human text, sketch or reference goals become planned brushstrokes, physically executed and revised from camera feedback.
Practical use, access and limits
Practical use: The artist supplies goals and material context. It demonstrates a genuine canvas output and the importance of adapting to actual paint behaviour.
Access: open paper/project and shared official robot code; hardware not tested
Limits: Requires robot, camera, calibrated workspace and material setup. A simulator or open repository does not provide a ready studio assistant.
Tags: physical-output · own-reference-style · robot-painting · material-feedback
CoFRIDA: Self-Supervised Fine-Tuning for Human-Robot Co-Painting
Peter Schaldenbrand and colleagues
A partially painted canvas and text goal guide new robot marks, with a model adapted to physical execution constraints.
Practical use, access and limits
Practical use: The person keeps painting and redirecting the collaboration. The project specifically documents human watercolour combined with robot marker work.
Access: open paper/project and shared official robot code; hardware not tested
Limits: Research robot integration requires calibration and compatible materials. Published mixed-media examples do not guarantee preservation of any artist’s existing marks.
Tags: physical-output · co-drawing · mixed-media · robot-painting
Diffusion Illusions: Hiding Images in Plain Sight
Ryan Burgert and colleagues
Chosen prompts or images guide jointly optimized images that reveal different content when flipped, rotated or physically overlaid.
Practical use, access and limits
Practical use: The artist chooses paired readings and arrangement. Offers an edition/object route through transparencies, backlighting or hinged tiles.
Access: open paper/project, official notebooks and printable examples; not executed
Limits: Illusion constraints can dominate the imagery. Physical alignment, printing and light need trials; examples establish no compatibility with a particular painting practice. Demonstrated overlays use printed overhead sheets and backlighting; transfer to spaced hand-painted glass and coloured wall light is untested.
Tags: physical-output · mixed-media · print · optical-illusion
TokenFlow: Consistent Diffusion Features for Consistent Video Editing
Michal Geyer, Omer Bar-Tal, Shai Bagon and Tali Dekel
A source video and target prompt become an edited sequence using cross-frame feature correspondence to support consistency.
Practical use, access and limits
Practical use: The artist authors source motion and edit intent. An approved filmed brush gesture could become a controlled moving-image study.
Access: open paper, project and official editing code; not executed
Limits: Reconstruction quality and the base edit matter; authors note possible decoder jitter. Temporal coherence does not guarantee exact colour, contour or texture retention.
Tags: coherent-video · artist-directed-editing · video-to-video · temporal-consistency
Understanding and Creating Art with AI: Review and Outlook
Eva Cetinic and James She
Reviews both artwork analysis and new-art creation, with a bibliography spanning image collections, aesthetics and generative methods.
Practical use, access and limits
Practical use: Use its reference map to locate distinct artistic questions before choosing a model.
Access: open paper
Limits: Abstract and source record checked; not a current installation manual or proof that every cited method works on these paintings.
Tags: survey · art analysis · generative methods
Designing Creative AI Partners with COFI
Jeba Rezwana and Mary Lou Maher
A framework for modelling turn-taking, contribution and communication in human–AI creative systems; the authors analyse 92 existing systems.
Practical use, access and limits
Practical use: Ask whether the artist can correct, reject, redirect and take turns with the system.
Access: open paper
Limits: Paper abstract and record checked; the framework is not a validated prediction of Iana's preferences. The authors limit the observed interaction clusters to their historical 92-system dataset.
Tags: co-creation · human control · evaluation
Painting Robot e-David — project and collaborations
University of Konstanz Visual Computing
Explains image abstraction into regions/strokes, a paint plan and camera-based feedback while a robot applies real paint. Includes collaborations with Liat Grayver.
Practical use, access and limits
Practical use: Study how material limitations, stroke planning and observation change the process of making a painting.
Access: Full indexed primary project text retrieved through the web reader on 3 November 2026 after an earlier restricted fetch and direct HTTP 404. A separate university collaboration article also opened. Current-origin availability remains uncertain.
Limits: Hardware research platform; no claim this is a downloadable turnkey studio tool. Components include AI-based semantic information and conventional control.
Tags: physical painting · robotics · feedback · artist collaboration
Wekinator examples
Rebecca Fiebrink and contributors
Input/output examples and teaching examples for interactive machine learning across programming environments.
Practical use, access and limits
Practical use: A starting point for learning a relationship between an artist's gestures and a visual or sonic response.
Access: free code repository
Limits: Repository contents checked; dependencies/examples were not installed or executed. Not an image-generation painting model.
Tags: gesture · interactive machine learning · teaching · code
Machine Learning Education for Artists, Musicians, and Other Creative Practitioners
Rebecca Fiebrink
Describes learning objectives, teaching strategies and evidence from courses for creative practitioners.
Practical use, access and limits
Practical use: Learn by building, demonstrating, testing and refining expressive systems, rather than beginning with model metrics alone.
Access: free accepted-manuscript PDF
Limits: Public PDF opened; historical course studies do not establish present course enrolment or current software compatibility.
Tags: education · artist learning · interactive systems
Picbreeder: A Case Study in Collaborative Evolutionary Exploration of Design Space
Jimmy Secretan and colleagues
Documents people selecting, evolving and branching images through collaborative interactive evolution and neuroevolution.
Practical use, access and limits
Practical use: An alternative to a single prompt: let artistic preferences guide a branching search through possibilities.
Access: free author-hosted manuscript PDF
Limits: Historical research, not diffusion. The original hosted interface was not tested; an available paper does not establish a working service.
Tags: human selection · evolutionary art · collaboration · historical
Projects: Technelegy, text works and media-rich poetry
Sasha Stiles
Documents AI co-authored poetry alongside text paintings, mixed-media objects and moving-image adaptations.
Practical use, access and limits
Practical use: See how an authored text corpus can feed a sustained practice across digital and physical publication.
Access: free artist project pages
Limits: Adjacent literary/visual practice, not an abstract-painting recipe. Some descriptions are artist statements, not reproducible training instructions.
Tags: poetry · artist archive · physical editions · moving image · adjacent
Association for Computational Creativity — publications and teaching resources
Association for Computational Creativity
A research hub linking conference proceedings, bibliographies, journals and teaching resources.
Practical use, access and limits
Practical use: Use its curated references to extend beyond a fixed list of tools or famous artists.
Access: free directory and proceedings links
Limits: A discovery portal rather than one production tutorial; homepage event text may be dated. Check individual papers and course availability.
Tags: bibliography · research archive · teaching · discovery
What has been checked.
“Opened” means the linked source page, abstract, publisher record or document was retrieved to support its annotation. It does not mean every book was read in full, every video watched or every model run. Book records and access restrictions are described individually.
A firsthand interview reports the artist’s own process; it is not an independent replication. Research projects may depend on historical software, missing datasets or specialist hardware. Some sources address neighbouring computational art, sound, poetry or upstream data practice; their limits are stated rather than treating everything as a tested painting workflow.
One historical course entry is marked search-only and closed for enrolment. Free publisher pages are not labelled as free full books. Where open access is advertised but retrieval failed, that distinction remains in the annotation.
All 67 records were reviewed again on 3 November 2026. The linked audit records corrections, partial full-text access and current-origin retrieval limits. Separate generated image studies received reported positive feedback for Meridian Light; structured artist-led evaluation and physical trials remain outstanding. This is not an exhaustive literature review.
Download the annotated bibliographyDownload the source tableDownload the research recordsRead the audit and three next experimentsDownload the source auditDownload the experiment briefs