ok
its the extracted lora from
so what doeas that lora do
for non distilled models you can use it with positiv weights ~ +0.5 or more
-> now you can set CFG from 1 to 2(or more) and you can use negprompt without getting orange blured images (but slower generation speed)
for distilled models you can use it to reverse that, so weights ~ -0.5 or more
and you can set CFG to 1, no negativ prompt but faster generaton speed
In some cases, I have found that the images are colder and sharper with positive weights and cozier with negative weights
Shutte reduces steps but its sharp and not natural, so use both loras and you get the images i post ;)
best with
Forg-FLUX(DPM++2Sa) and Beta
Please play with it ;)
Description
First Version
FAQ
Comments (2)
There is something I don't understand. I already use DE-DISTILLED Flux for everything. What improves using this LoRA?
so DE-distilled is distilled with neg-prompt option (cfg above 1) ?
than maybe you dont need it ...
BTW i hate FLUX for that 10 base versions and nobody knows which is what ^^
hyper/step/distill/fp8/fp16/nf4/gguf/ with and without unet/clip/t5xxl ?!?!?!?!?!?!?
