A five-year study of Chinese digital painters offers a rare view of generative artificial intelligence as a lived creative problem rather than a one-day product announcement. The researchers followed 17 artists from 2021 through 2025, returning annually to ask which parts of practice they would protect, which they might delegate, and where they still located authorship.
The result is not a simple story of rejection turning into acceptance. It is a record of moving boundaries. Some artists continued to refuse AI in finished work. Some used it for references, backgrounds, rough sketches, or client-facing drafts while reserving composition, faces, final passes, and signature decisions for themselves. Others described fatigue, pressure, or difficulty identifying what remained distinctly human when clients and production schedules rewarded speed.
From resistance to bounded use
The study, published as a September 2026 preprint by Yibo Meng, Ruiqi Chen, Shuheng Cao, Weijia Zhang, and Chengxi Zang, is based on annual semi-structured interviews conducted between 2021 and 2025. The baseline matters: in 2021, participants were encountering a different technical landscape, dominated by style-transfer tools, avatar generators, and early image-generation experiments. The later interviews followed the spread of prompt-controlled systems into public culture and professional pipelines.
The authors describe three recurring, but non-uniform, patterns. Early protective resistance defended observation, originality, signature, and ownership. Later, pragmatic delegation appeared in bounded tasks. By 2025, some participants were actively revising those arrangements, deciding again what a tool could do without inheriting the artist’s authority.
The useful phrase is “agency partitioning.” It names the practical act of assigning tasks, responsibility, value, and credit. An artist may let a system suggest backgrounds but not compose the central figure. Another may use generated references but refuse generated material in a delivered image. A third may accept AI for color exploration while keeping the final mark and signature human. The boundary is not defined once; it can change as tools, clients, and peer expectations change.
Efficiency is not the only measure
Production pressure runs through the findings. Eleven of the original participants were professional digital painters working in fields including games, publishing, and design. In fast pipelines, a tool that shortens a task can be difficult to ignore. Yet faster production can also reset client expectations, reduce fees, or make a formerly skilled stage feel invisible.
That tension helps explain why artists did not evaluate AI only by image quality. They also asked who observed, who intended, who remained responsible for errors, whose labor was being valued, and whether a finished image could still carry a credible claim of authorship. Copyright and training-data concerns remained recurring limits even when participants found particular uses convenient.
Peer communities mattered too. The same use could feel acceptable in one professional circle and discrediting in another. Artists were not only negotiating with software; they were negotiating with colleagues, clients, audiences, and their own earlier positions.
A practical studio exercise
The study suggests a more useful question than “Do you use AI?” Artists can map their process stage by stage: research, references, sketching, composition, color, texture, background, revision, final rendering, delivery, documentation, and promotion. For each stage, they can decide whether AI use is prohibited, experimental, permitted with review, or acceptable for final production.
That map should also identify who remains accountable. If a generated reference introduces an unverified object, a copied motif, or a misleading visual claim, the artist or studio still needs a check. If a client requires disclosure, the workflow should preserve prompts, source files, and revision history. If a collaborator contributes to a hybrid work, credit should be agreed before delivery rather than reconstructed afterward.
A human-only zone can be small or extensive, but it should be deliberate. It might protect direct observation, the initial drawing, a signature material process, the final edit, or the decision about what the work means. The study’s participants show that these zones are not nostalgic barriers. They are working definitions of responsibility and identity.
What the research cannot settle
The authors are careful about scope. This is a qualitative preprint based on a small cohort in one national and professional context, and interview availability declined across the five waves. It cannot establish how all visual artists are changing their practice or which workflow is best.
Its value is different. Because the same people were interviewed repeatedly, the research captures revision: curiosity becoming resistance, resistance becoming bounded use, and efficiency later being reinterpreted as pressure or compromise. That temporal view is often missing from surveys taken at a single moment.
For artists, the central lesson is that creative agency does not disappear in one dramatic transfer. It is redistributed through many small choices. Naming those choices makes it easier to protect authorship, explain a process to clients, and use—or refuse—a tool on terms that remain legible.
ATIM approaches artificial intelligence with curiosity, independence, and balance. We explore its extraordinary potential to empower artists while examining the questions it raises around authorship, copyright, consent, creative identity, transparency, and human expression.
We are neither advocates for AI nor opponents of it. Our focus is how the technology is used—and what that means for artists.
Above all, our commitment is to artists. We look at every development through the lens of what can support, protect, and empower the creative community—identifying opportunities, examining potential risks, and providing artists with the information they need to make informed decisions about their work, their rights, and their future.