So I can reliably get Flux to replicate my signature.
This is interesting, and it explains a lot about the model.
Because, if I'm in there, of all people, it means they intentionally trained the thing on adversarial inputs. In other words, art that's designed intentionally to break the denoising process that's typically used in training and rendering on diffusion based systems.
I suspect they're using pieces of my artstation portfolio from over a year ago, which is fine. But things didn't really get interesting with noise painting until 2024.
It also means that the training process spent enough time with my signature, and my art to replicate it successfully. This makes me happy.
And that explains (in some small part) why it treats line work the way it does.
Also, it feels good to know that, in some way, I have influenced the machine, and everything that will ever come from it!
These models are historic documents. Being in them now, matters, if what you are about is your influence over the future of art. It feels awesome to be included in art history like this.
I am honored.