How to keep your art out of AI generators

Generative AI is everywhere you look, but for working visual artists, the boom often feels like watching their livelihoods get scrapped for parts. Generative models routinely hoover up public web art without consent or compensation, and simply deleting your social presence isn't an option when commissions depend on portfolio visibility.

Navigating the current protections is messy, but visual creatives do have a few levers left to pull.

The ordeal of the official opt-out

Most AI providers design their opt-out forms to be just compliant enough to ward off regulators, but tedious enough to discourage anyone from actually using them. OpenAI lets creators request to exclude their work from future DALL-E 3 training, but there is a brutal catch: you have to fill out an individual form for every single image, complete with a file upload, written description, and rights confirmation. If you have a portfolio with thousands of illustrations, good luck. You can block OpenAI's GPTBot on your personal website via robots.txt, but that won't stop copies posted on social platforms from being scraped, nor will it purge anything already baked into existing checkpoints.

Other big platforms are just as restrictive. Adobe markets Firefly as legally safe because it trains exclusively on Adobe Stock. However, Adobe Stock contributors cannot opt out of AI training at all—if you want your art off Firefly, you have to leave the marketplace entirely. Meta scrapes public posts across Facebook, Instagram, and Threads by default. While Meta offers a personal data request form, artists say it frequently fails; conceptual artist Bethany Berg told Wired it felt like a "fake PR stunt."

Meanwhile, platforms like Midjourney and Stability AI offer no forward-looking opt-out mechanisms whatsoever. Your only recourse there is manually firing off DMCA takedown requests to email addresses buried deep in their terms of service.

Cloaking images before uploading

Because corporate opt-out forms are largely passive-aggressive roadblocks, researchers have stepped in with adversarial tech. The most prominent is Glaze, a free tool developed by the University of Chicago.

Glaze applies imperceptible pixel-level cloaks to artwork. While human viewers see the painting normally, generative models reading the underlying data interpret the image as an entirely different art style. If an unauthorized scraper feeds the image into an AI model, the model fails to learn the artist's real stylistic patterns.

Why it matters

We think the current opt-out landscape puts an unfair burden on artists. Companies have made the friction of protecting one's work absurdly high, turning basic copyright defense into an unpaid full-time administrative job. For now, visual creators are much better off using active defense tools like Glaze or keeping full-resolution files locked behind private links, rather than wasting days submitting one-by-one forms to OpenAI or Meta.

This dynamic also exposes how obsolete legacy copyright tools like the DMCA are in the generative era. DMCA takedowns were built to remove hosted copies of pirated files, not to regulate model weights that absorb artistic style without storing raw pixels. Until regulatory frameworks require explicit opt-in consent for machine learning datasets, we expect adversarial poisoning tools to remain the creative community's only real shield.