Adobe announced Firefly, its first model covering image generation and text effects. What distinguished it was not the output but the provenance of the training data: only Adobe Stock, the company's own royalty-free library, together with openly licensed and public-domain material whose copyright had expired. The pitch was that what you made with it was safe to use commercially. Adobe said it was working on a way to compensate Stock contributors and would let creators attach a "do not train" credential to exclude their work. Coming weeks after a run of lawsuits over training data, it argued for a generative model's legitimacy through sourcing rather than capability — a framing others would borrow.