Update README.md

This commit is contained in:
Extraltodeus
2024-06-20 17:37:16 +02:00
committed by GitHub
parent 7bd9cfd020
commit 5db56ae11c
+1 -1
View File
@@ -3,7 +3,7 @@ Allows to sample without generating any uncond with Stable Diffusion!
This **doubles the generation speed** while removing the control provided by the negative prompt. Yet still preserves image quality.
Subtracting the mean value of the latent space and the positive prediction as well as using an adapted scale is basically what allows this. Additionally, using a "good model" (base models seems to be terrible in these conditions) as well as a recent sampler such as dpm++2m sde or quite the majority of [Clybius Extra-Samplers](https://github.com/Clybius/ComfyUI-Extra-Samplers) helps a lot!
Subtracting the mean value of the latent space and the positive prediction as well as using an adapted scale is basically what allows this. Additionally, using a "good model" (base models seems to be terrible in these conditions) and a recent sampler such as dpm++2m sde or quite the majority of [Clybius Extra-Samplers](https://github.com/Clybius/ComfyUI-Extra-Samplers) helps a lot!
Note: This is a work in progress and may go through some changes!