Introduction from Alimama
【This checkpoint is a 8-step distilled Lora, trained based on FLUX.1-dev model. We use a multi-head discriminator to improve the distill quality. Our model can be used for T2I, inpainting controlnet and other FLUX related models. The recommended guidance_scale=3.5 and lora_scale=1. Our Lower steps version will release later.】
And this is PAseer-Optimazed Version
Less influnce of Asian character trend than Alimama.
Avoiding adding extra things which not in prompts.
Less Storage Volume Usage-- 150 MB
Description
FAQ
Comments (9)
This lora at 0.8 has a huge impact to face.
This is amazing! I'm not using it to reduce steps (which it helps with well of course), but I'm using it as a detailer lora. It's incredible! It works well with other loras and from my initial testing I think it even helps with prompt adherence to a degree. I'm using it with 35 steps and the images are amazing in regards to sharpness and texture detail. So far I haven't noticed any significant decrease in regards to anatomy, hands, fingers or such. I'm impressed.
Time flies faster with it. Flux is so sloooow... Zzzz ;-)
https://github.com/welltop-cn/ComfyUI-TeaCache
Try this。 24G GPU can speed up to 5 its/s.
@Aseer Ho very interesting! Thx ! I have a super 16G 4080
Doesnt work in 'chroma GUFF'-like checkpoints, like chromaGGUF_v28Q6K.
Bit late, but majority of LORAs work really badly with GGUFed Chroma. I think its not GGUFed correctly.
One of the best LoRA's for flux, better for realistic portraits than every other turbo I've tried (5-6 others) as far as convergence without degrading quality.
Worse than the original one
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.