2025Design Frameworks, UX Research, AI Ethics
Artists vs AI
Ten Instagram artists on what generative AI took, and what would give it back.
Small artists are losing visibility, credit, and income to generative models trained on work like theirs. We ran contextual research with ten of them and turned 202 interpretation notes into design directions a platform could act on.
- Role
- UX Researcher
- Team
- Soyon Kim, Erica Flora Yu, Esther Suh, Elise Zur, Christy Yu
- Method
- Contextual interviews and directed storytelling
- Sample
- 10 Instagram artists under 10k followers
- Links
- Read the zine
Artists are buried under work trained on their own
Original work by small Instagram artists is being buried under generative content, some of it trained on that same work. The artists we spoke to were spending real energy protecting an artistic identity that the platform gives them almost no tools to protect.
Memories, not opinions: 202 notes from ten artists
We were after the things artists feel but do not post about, so we asked for specific memories rather than opinions.
Contextual interviews
45 to 60 minute interviews to understand how artists feel about generative AI in the situations where they actually meet it. Participants were recruited through personal networks in the artist community.
Ten artists
Ten small Instagram artists under 10k followers, all consistently active over the previous six months. Illustrators, digital painters, animators, and mixed-media artists.
Directed storytelling
We prompted participants to recall specific moments involving generative AI. Anchoring on real events surfaced honest emotion and real decisions instead of rehearsed positions.
The interviews produced 202 interpretation notes. We synthesised them through affinity clustering, then built an empathy map and a cultural model so we were reading the data from more than one angle.
Self-protection has a ceiling, so the platform has to act
Five quotes, and what each one asks a platform to do.
I struggle to tell, and you have to start zooming in to see the details, and then you feel bad, because what if it has been made by real artists? It is a bit disheartening.
Artists are running manual forensics on their own feed. Disclosure should be automatic and system-side: platforms and AI tools should say when content is generated or trained on a specific artist's work, so the burden of policing infringement stops sitting on the people being infringed.
People think of AI with black and white thinking, but there is the gray area as well. I do not know if I have a platform to share this opinion.
Several artists hold more complex private views than they can say publicly. Many see practical uses for AI in ideation or admin work and fear the social cost of admitting it. The polarised discourse pushes people into camps they do not fully live in, and it blocks the conversation about responsible integration before it starts.
Humans are responsible for making sure AI does not take over artistic standards.
Artists locate the responsibility with people and institutions, not with the technology. That is an argument for explicit platform-wide rules: what disclosure is expected, what use is permitted, and what happens when the rules are broken. Without them, the standard defaults to whatever is easiest to publish.
You can see what used to be a signature or what used to be a watermark in the corner of an image, that sort of proves it came from another artist.
One artist described watermarks as a way to catch generated work, because models imitate the signatures of the humans they learned from. Individual defences like this leave artists feeling they have lost agency entirely. Making AI origin visible, enforceable, and simple to filter is a systemic answer to a problem people are currently solving alone.
With the age of AI, I need to argue more to people who are unable to understand the beauty of human-crafted products, ideas, and artwork.
Artists lit up talking about the small behind-the-scenes details of their work, almost as proof of authorship. The emotional connection to a piece thins out when the effort behind it is invisible, which makes process visibility a design problem and not just a marketing one.
Five directions, from disclosure to co-designing with artists
Five directions a platform could take, ordered from what it owes artists to what it could build with them.
- Invest in transparency features that state plainly how models interact with an artist's content, including no training and opt-in collection.
- Build creator protection tools: style protection, watermarking, and blocking model scraping.
- Pilot AI features with small artist groups to co-design ethical creative tools around ideation and moodboarding.
- Define clear platform-wide guidelines for AI use, including disclosure expectations and what is not allowed.
- Support community-led initiatives that make human-made art, and the process behind it, more visible.
