model_patches/Automatic_CFG/experimental_attention_modifiers
MODEL
My own version 'from scratch' of a self-rescaling CFG. It isn't much but it's honest work. TLDR: set your CFG at 8 to try it. No burned images and artifacts anymore. CFG is also a bit more sensitive because it's a proportion around 8. Low scale like 4 also gives really nice results since your CFG is not the CFG anymore. Also in general even with relatively low settings it seems to improve the quality.
Authored by Extraltodeus
No downloads or installs are required. Pay only for active GPU usage, not idle time. No complex setups and dependency issues
Learn more