How do you figure? Because they say so? Anyway, where do they say it, for Firefly? I haven’t seen it. They don’t address the text encoder at all, which uses images, precisely because it’s something like T5 or CLIP. I’ve emailed them through my enterprise Adobe account about it, and they simply didn’t reply. And then suppose they did talk about this, who audited it? How do you know they’re not biding time? Those text encoders take forever to train.

At the end of the day, we’re only talking about these models because someone published research completely tainted by the copyrighted training. So it’s still totally unknown, until a lawsuit, if even a “redo” of every part of the model on all licensed stuff, which may not even work because there isn’t enough variety, if it’s all that it promises legally.