What happened with Stable Diffusion and artists on December 14, 2022?

Spawning, a group building tools for artists, announced that Stability AI would honor requests from artists to keep their work out of training for the next version of Stable Diffusion. Artists could register through the site haveibeentrained.com.

Was the opt-out the same as asking permission first?

No. Artists had to find their work and remove it themselves, and critics quoted by MIT Technology Review argued that the default should have been opt-in instead.

How should I prompt for an art style without leaning on an artist's name?

Describe what you actually like about the look, such as the palette, the line weight, the lighting and the mood. A description gives you more control than a name does.

Dec 14, 2022 · Artists get a Stable Diffusion opt-out

The day artists got a way to say no to an image model

By R.Solace · Written November 1, 2026

🤖 AI-assisted, human-curated. How we use AI →

Two weeks after ChatGPT arrived, most of the noise was about chatbots writing essays. The fight that mattered to a lot of working illustrators was somewhere else. It was about pictures, and about whose pictures went into the machine in the first place.

On December 14, 2022, Spawning announced that Stability AI would honor artists' requests to opt out of training for the next version of Stable Diffusion. VentureBeat reported it the following morning and dated the announcement to the day before.

What actually changed

Spawning was a young outfit. VentureBeat says it launched in September 2022 to build tools that give artists ownership over their training data, and it was co-founded by Mat Dryhurst and the musician Holly Herndon. Its site, haveibeentrained.com, let you search the image set a model learned from and flag your own work.

The deal was that flagged work would be left out of future training. The opt-outs were expected to start within a few weeks. MIT Technology Review, reporting two days later, said Dryhurst put the window at roughly a couple of weeks before training on the next model began.

Stable Diffusion had been built on LAION-5B, a dataset scraped from the internet that included copyrighted images. So for a lot of artists this was the first time anyone had offered a button at all.

Why it didn't settle anything

My take: an opt-out was better than nothing, and also clearly not enough. MIT Technology Review quoted Karla Ortiz, a board member of the Concept Art Association, arguing that the approach put the burden on artists, who had to opt out rather than being asked to opt in. Her group was raising money to hire a lobbyist in Washington.

Stability's chief executive, Emad Mostaque, said on Twitter that the company didn't believe it had a legal reason to do this, and framed it as a chance to compare what models trained on different datasets would produce. VentureBeat also noted that Dryhurst himself admitted enforcement couldn't be complete, since anyone could scrape the web.

So, a precedent and a fight, both at once. The fight is still going.

What this means for your prompts

You can't change what a model was trained on. You can change how you ask it for a look.

R.Solace

R.Solace writes about prompts the way a shop owner talks about orders: what you ask for is what you get.

Try free prompts →The full timeline →

The day artists got a way to say no to an image model

What actually changed

Why it didn't settle anything

What this means for your prompts

Try it

Quick answers

Sources