Comment by pjc50

2 days ago

Wait, how exactly did Adobe create noninfringing models?

Edit: not a full explanation, but https://www.mikechambers.com/blog/post/2025-09-24-generative... ; this is subtly different. It's a claim that the model will not create infringing output, but that's not the same as "this model was trained only on content which was licensed for the purpose of AI training".

(there's also a discussion of the idea that the output of a model may not be copyrightable at all, which will cause a whole second set of problems for commercial users)

1 comment

pjc50

ben_w 1 day ago

According to them:

  Adobe Firefly models are trained on a dataset of licensed content, such as Adobe Stock, and public domain content where copyright has expired. Adobe Stock content is covered under a separate license agreement, and Adobe compensates contributors for the use of that content.

  We do not mine the web or video hosting sites for content. We only train on content where we have rights or permission to do so.

- Under "Our Approach.", all of which starts pre-collapsed (why is this a thing?): https://www.adobe.com/ai/overview/firefly/gen-ai-approach.ht...