This was already addressed from the beginning: https://opengameart.org/content/artificial-intelligence-assisted-artwork

"Note: Submissions will be unaffected by any such removal if the art therein was generated with technologies who's training dataset, in its entirety, is demonstrably openly licensed."

I think it depends on what the "those with open licenses" are - I mean, if it was trained on a million CC BY images, and you need to abide by those licences, that's up to a million attributions to list...

^^Agreed. Thus the "demonstrably". Links to Adobe Firefly dataset information?

i did some perusing, and there is no transparency of the dataset that i can see. "adobe says" is the only think i can see.

from the FAQ 

"

The current Firefly generative AI model is trained on a dataset of Adobe Stock, along with openly licensed work and public domain content where copyright has expired.  

As Firefly evolves, Adobe is exploring ways for creators to be able to train the machine learning model with their own assets so they can generate content that matches their unique style, branding, and design language without the influence of other creators’ content. Adobe will continue to listen to and work with the creative community to address future developments to the Firefly training models.

With the Content Authenticity Initiative (CAI), Adobe is setting the industry standard for responsible generative AI. With more than 900 members today, the CAI is leading the conversation around digital content attribution. The CAI offers free, publicly available open source tools and is working on a widely adopted technical standard in collaboration with leading technology organizations through the nonprofit Coalition for Content Provenance and Authenticity (C2PA).

Firefly will automatically attach a tag in the embedded Content Credentials to make AI-generated art and content easily distinguishable from work created without generative AI.

"

however i cant find any "demonstration" of there claims. i mean they obviously own the rights to adobe stock images, but i guess we are just supposed to trust them that the rest of the dataset is public domain.

but since my first generation with it churned out this thing, it cant be all bad:

I would say, in this case, "Adobe says" is actually sufficient. They are publicly asserting (which makes them legally liable) that all assets used to train Firefly are correctly licensed/attributed already or they posess all relevant rights.

So we will be allowing art created by Firefly on OGA?

I'm open to hearing arguments to the contrary, but barring that, yeah. :)

EDIT: missed the fact that Firefly output is (currently) non-commercial only. Until that changes, we can't host Firefly-generated assets.

under current terms i believe the content made using firefly is allowed for non-commercial uses, when i fiddled with it it applied a no commercial use watermark to my images (see the attatchment above)

i do think it is really cool that companies are paying attention to the debate and trying to do this in a way that doesn't violate anyone's rights.

i think in the future we will see more open datasets

Oh, dang. I totally missed that. Yeah, unless Firefly has a non-non-commercial option, that won't do.