Hacker News Reader

Comment by Manuel_D | Hacker News Reader

But the images that went into building the models are copyrightable. Adobe has a license to use them for AI training, but the authors still have copyrighht.

So as per your logic the answer would actually be "no, these images aren't exempt from copyright."

View on HN

Nothing from the OP or the original AI Artist was about using Adobe.

That said, were the images used by Adobe actually licensed for AI training, or was an existing overly broad license aimed at "showing this in an online gallery" interpreted by lawyers to include AI training? Because that latter option is being done by a lot of corporations (including Google) to make "ethical" AI models.

Adobe trains its models on Adobe Stock and public domain images

> Adobe does not train Firefly Gen AI models on customer content. Firefly generative AI models are trained on a dataset of licensed content, such as Adobe Stock, and public domain content where copyright has expired.

https://slate.com/technology/2024/06/adobe-terms-use-backlas...

> Nothing from the OP or the original AI Artist was about using Adobe.

The point is that even using basic photoshop tools is "AI art".

Thank you for the info. It's still going to run afoul of the copyright office's interpretation of who owns copyright for AI generated materials, but I'm really glad to see them doing something ethical with the model creation.

So your understanding of the law is that any use of the inpainting tool in modern versions of Photoshop makes the image impossible to copyright? Heck, even the magic-wand select is using AI to fine tune its behavior.

Not the entire image, but parts created by inpainting. Which is admittedly a hot mess of red tape in potentia. But it makes sense from the point of view of copyright having been created to protect people's ability to make a living from IP.

The magic wand is different IMO, because it's not creating any imagery.

If you use the magic wand, and then do anything with the selection mask - cropping, changing hue, etc. - then you're creating imagery with AI.

If any if the training set is public domain or creative commons, I'd argue that this make the image at best an integration work, not derivative nor fair use, and as bits from the images are indistinguishable from each other, you shouldn't be able to license it.

Like code generated from models trained on AGPL/GPLv3 licensed code should also be AGPL/GPLv3.

Reply on news.ycombinator.com