Help has arrived

tal@lemmy.today · edit-2 12 days ago

Help has arrived

tal@lemmy.today · edit-2 11 days ago

One thing that I like about some Stable Diffusion-derived models is that they have been trained on the works of many artists, can replicate styles.

Flux, while rather more-sophisticated in many ways, including the ability to use natural-language image descriptions, has – probably intentionally – not been. And while I really like Flux’s functionality, I’ve been kind of unhappy about losing that.

However, there are models derived from Flux – as the one used here – that have had that trained back in.

EDIT: Note that this model still can’t do some of the things that I’ve done in the past with Stable Diffusion-derived models trained on a lot of artworks, like these landscape paintings in the style of Casilear, but the fact that it can do something at all is a considerable improvement over the dev version of vanilla Flux. I did read one comment that the stable version of Flux supports artist names, but then you’re stuck using their service to do your generation; the stable model of Flux isn’t distributed.

Even_Adder@lemmy.dbzer0.com · 11 days ago

You’re killing it with these gens.

tal@lemmy.today · 11 days ago

Thanks, though only “Cats” was really just put up for the looks; the other three were to show off some functionality to other folks who might be doing local generation themselves, to let them know about functionality that I’d run into that I wanted and give 'em a heads-up as to it being possible (a way to get decent food photography, a way to emulate artist styles again under Flux, and that the ability to generate progressive transformations in Flux exists).