The text-to-image generating service DreamUp, powered by Stable Diffusion and AI, was introduced by the online art community DeviantArt on Friday. At the same time, DeviantArt introduced a programme that purportedly allowed artists to opt out of AI image training but really made everyone’s art opt in by default, upsetting many users.
DreamUp uses language prompts to generate original AI-generated art. Due to the Stable Diffusion foundations of DreamUp, some DeviantArt users may find it ironic that DreamUp learned how to create images by studying hundreds of millions of images taken from websites like DeviantArt and aggregated into LAION datasets without artists’ consent.
this is how dreamup is being developed. it adds training on non-opt-out artwork on top of the stability diffusion model, which is ALREADY trained using unethically sourced data. they CANT UNLEARN those parts.
this whole discussion is not in good faith unless they address this pic.twitter.com/KnQZerCIol
— svlt ✈️ (@svltart) November 11, 2022
Stable Diffusion’s web-scraping capabilities, about which we’ve written a lot on Ars in the past, sparked a heated discussion earlier this year among artists who question the morality of AI-generated art. Some art societies have firmly opposed AI-generated images and outright forbidden them.
DeviantArt is attempting to appease artists who might be angry about their work being used to train AI image generators, possibly anticipating a backlash. Artists can choose not to use third-party image datasets by selecting the special “noai” flag provided by the website in their image settings. (However, it is unknown whether third-party image scrapers will respect this flag.)
Additionally, DeviantArt will provide users the option to prevent their photographs from being used to train DreamUp in the future, but each user must first complete a form that needs to be reviewed by a human. Members of DeviantArt have strongly reacted against this decision, with some threatening to erase all of their work and deactivate their accounts.
Heads up, DeviantArt turned heel and as far as i can tell all of your existing deviations will have to be manually opted out of AI data sets.
YOU ARE OPTED IN AT DEFAULT
— ɪʀɪꜱʜ ᴍᴇɴᴀᴄᴇ I Λ П | F Λ Y (@IanFayArt) November 11, 2022
The DreamUp information page for DeviantArt likewise adopts a defensive stance, claiming that DeviantArt did not authorize third-party AI image models (such as Stable Diffusion) to scrape their website in order to create their models. The website also makes an effort to dispel many misconceptions about how AI image synthesis functions farther down the page.
In response to community input, all photos hosted by DeviantArt will be “automatically designated as NOT permitted for usage in AI datasets,” according to an update to DeviantArt’s AI dataset policy.
This now only seems to apply to third-party AI datasets that carry the “noai” tag, which dataset scrapers will have to actively respect in order to be effective (we know of no plans to honor this flag at this time). Artists will still need to complete a form if they do not want their work to be used in DeviantArt’s own AI training for DreamUp.