Artists gain a way to exclude work from Stable Diffusion 3

Stability.AI said artists can use Spawning’s HaveIBeenTrained website to select works they want excluded from the training data for the next version of Stable Diffusion. The opt-out plan has drawn criticism from artists who argue that inclusion should require consent.

WTF Index TERMINATOR
◄ Terminator 2 Idiocracy 1 ►

The opt-out offers artists some control, but the story centers on AI training with scraped copyrighted work and consent concerns.

Artists gain a way to exclude work from Stable Diffusion 3

Artists will be able to identify works they want excluded from training data for the next version of Stable Diffusion. Stability.AI announced the plan with Spawning, whose HaveIBeenTrained website lets artists search for their work in the data set used to train the image generator.

How the opt-out is meant to work

Artists can search the training data through HaveIBeenTrained and select which works to exclude. Spawning was founded by artist couple Mat Dryhurst and Holly Herndon. The service is intended to give artists a way to make choices about whether their work is used to train the next model.

Dryhurst told MIT Technology Review that artists had “around a couple of weeks” to opt out before Stability.AI began training Stable Diffusion 3. The company described the change as part of work with collaborators to improve its data set, including options for people to opt out and, when not already included, opt in.

Stable Diffusion is based on the open-source LAION-5B data set, which was built by scraping images from the internet, including copyrighted artists’ works. Some artists’ names and styles have also become popular prompts for people generating images with AI. That background has fueled disagreement over what permission should look like when creative work is used in model training.

Why artists disagree with the approach

Karla Ortiz, an artist and board member of the Concept Art Association, says an opt-out still assumes that artists are included unless they take action. She argues that this removes artists’ choice by default.

Ortiz says that Stability.AI should use “algorithmic disgorgement,” which she describes as destroying the database and models containing artists’ data. Her position reflects a broader dispute about whether a searchable opt-out can adequately address the use of copyrighted work in training data.

Dryhurst sees the opt-out service as a practical measure while the industry lacks clear standards or regulation around AI art and intellectual property. He also said that artists could eventually opt in to having their works included in data sets. The proposal therefore points toward both removal and participation, though the immediate focus is on allowing artists to select work for exclusion.

A broader fight over creative work

The Concept Art Association is raising $270,000 to hire a full-time lobbyist in Washington, DC. It wants changes to US copyright, data privacy, and labor laws that it says would protect artists’ intellectual property and jobs.

The group’s aims include updating intellectual property and data privacy laws to account for new AI technologies, requiring AI companies to follow a strict code of ethics, and working with labor unions and industry groups involved in creative work. These goals extend the debate beyond a single model or dataset to the rules governing how AI tools affect creative industries.

Ortiz says the association is warning other creative fields that AI tools are approaching creative professions “really fast,” and that the current approach is “extremely exploitative.” For artists, the opt-out offer may provide a concrete way to flag individual works, while questions about consent, accountability, and the protection of creative jobs remain part of the larger argument.