unStable Revolution F2K
Flux 2 klein, improved female anatomy
early alpha verion
Recommended Settings:
Sampler Euler / DPM++ 2M SDE
Scheduler Simple / Beta
4-5 Steps
Use any resolution, i like 896x1152 the most
Not tested with negative Prompts
Description
FAQ
Comments (6)
Thank you for this. Very rare to find an edit model that can run on 8gb vram. If you like, if you do any other releases of this in future, you might include a 4km gguf along side q8_0. Would likely enable slightly higher resolution without much loss in precision. Anyway, thanks! 🙏
@lauraknopek8329 i will publish a q4
@Peli86 A q5 is probably fine (you published after I posted). But nice to have options, anyway!
@lauraknopek8329 q5 is already ver small, guess a4 will lose stability. bit i'll give it a try. it's an experimental model, so there is nothing to loose ^^
can it run on 6 gb v ram
@rajaditya4033558 For the q5, yes you probably can! Bear in mind you need qwen4b in some quant, and a vae to run it (safetensors doesn't have everything). So like the vae is small and the safetensors is like 2.9gb ish so plenty of room there, but you need something that can offload things selectively, so that the text encoder (LLM in this case, q4 quant is about 2.5gb, a bit tight to hold in vram full time along side the unet weights) runs, and then gets out of vram for the model.