Stable Diffusion draws controversy on Twitter from artists who say the AI infringes on copyrights

Stability.ai's beta release of its text-to-image software has less filtering than DALL-E 2, and raises new questions about how it mimics other artists

A collection of modern or currently working artists from https://proximacentaurib.notion.site/e2537cbf42c34b7e9a9a4126f81dfd0d?v=7b4a3c03fb654045be324eb71acc57e6

Artists are embroiled in a heated debate on Twitter today about whether a new text-to-image AI called Stable Diffusion infringes on the rights of artists.

RJ Palmer @arvalis

A new AI image generator appears to be capable of making art that looks 100% human made. As an artist I am extremely concerned.

Stable Diffusion was released into beta this past week by startup called stability.ai. Unlike DALL-E 2, it doesn’t filter public figures, and it is trained on a dataset called LAION Aesthetics, which contains 120M aesthetic samples of image-text pairs that contain aesthetically pleasing images. A sample of those images is available here, and the Github repo is available here.

“What makes this AI different is that it’s explicitly trained on current working artists… This thing wants our jobs, its [sic] actively anti-artist,” said RJ Palmer, a concept artist who draws creatures.

Other artists weighed in on the debate, saying that the new AI is “morally abhorrent” and could even be considered theft:

Unsidhe | Vartist & Rigger| @ Othercon! @frostiefey

@arvalis I would consider this art theft. It's trained on the work of real artists. Real artists who are neither paid or credited. These artists should sue.

Joaquin Baldwin, a director at Disney, noted that since Stable Diffusion is going to be fully open-sourced, its advancement will be hard to prevent.

Joaquin Baldwin @joabaldwin

@arvalis SD is also going to be fully open, so no matter what, someone somewhere will have their own local copy with whatever dataset they want. Even if an artist asks not to be included in X site, there could be hundreds more popping up using their images in their training datasets.

No doubt, the technology also raises new questions about art education:

Kikidoodle & Purrmaids & Too Many Cats?!? @KikiDoodleTweet

@arvalis Things that are about to get really interesting (in a bad way): Reviewing portfolios to hire artists Professors grading assignments Any portfolio based review team

As of Sunday afternoon PST, the tweet thread from @arvalis had more than 8K retweets and 64K likes on Twitter.