[ideas] - retraining StableDiffusion #301
Description
to my surprise, although the Stable Diffusion model is incredibly expensive to train , apparently people are able to tweak it with a bit of retraining, e.g. one person managed to teach it about their own likeness.
I've been using the plain SD model locally myself and it's absolutely incredible to have,
Now I'm wondering if the polygon Level annotations that we have can be used to fine tune Stable Diffusion , e.g. make it better at texturing street scenes with the road/pavemnt annotations, help out with any gaps in what it's producing
the whole generative art phenomenon that's all the rage now is basically exactly why I took interest in your project in the first place.. it's interesting to watch the debate over legality, and all it takes to solve is for people to put time into open datasets, exactly like you tried to get going with this.. generative art needn't use scraped artist work, it should be possible with conceptual blending of natural and real world elements