Curator

(Points of clarification for this discussion;

1) We consider 'generative art' a separate entity to 'AI art', and generative art is totally fine and allowed on .art and not subject to this discussion;

2) This only applies to users on mastodon.art . If you're not on mastodon.art, this is irrelevant to you and won't affect you at all; refer to your own instance's rules.)

The only reason we haven't outright prohibited AI images in our rules until now was that barely anyone was posting it, and when they did, they'd usually get booed off stage – the community stance on AI images is very strong and obvious here, and that was enough to discourage people from posting it, or to educate themselves on why it's problematic.

Behind the scenes I've been rejecting applications from people who specify they want an account to post AI images, I never boost or interact with any AI images from Curator, and tend to mute off-instance accounts I see posting AI images using the art hashtags I follow just to keep them out of my search results. But just for us to have a more definitive stance on it, we're updating our rules to be more strict on what we're allowing going forward.

The main point of contention with AI-generated imagery is that the publicly accessible models people use to generate images were trained using other people's artwork without their consent. It's a glaring issue in ethics, and obviously one that's close to home for an instance full of artists who've had their work scraped by crawlers that pass on the data to things like Stable Diffusion for training.

As such, from this point forward, no AI-generated imagery is allowed on .art. If you want to post AI images, there are plenty of other instances that don't regulate posting it, and you can create an account on one of those.

We're aware that there are some grey areas, like people who generate and then heavily modify AI imagery to use in work that's otherwise their own and where those lines of ownership might begin/end, or people training AI on their own art – but given how prominent the ethics issue is with the training models right now, and the lack of transparency from a lot of models on where they got their training data, we're keeping the rules tight and all-encompassing. If, in future, some models emerge that are verifiably trained ethically without infringement on existing works, we will re-visit the rule.