Question about LORA - Githubissues

pamparamm / sd-perturbed-attention

Perturbed-Attention Guidance and Smoothed Energy Guidance for ComfyUI and SD Forge

MIT License

213 stars 14 forks source link

Closed alonsojr1980 closed 6 months ago

alonsojr1980 commented 6 months ago

Sorry, this is not an issue, just a question: connect the PAG's model input to the CheckpointLoader or to the LORA's output? Thanks.

pamparamm commented 6 months ago

~~Doesn't matter - output should be the same~~ Tested a bit, seems like PAG node should be connected after applying LoRAs

alonsojr1980 commented 6 months ago

Connecting after the LORA, the PAG's scale must be lower, right?

pamparamm commented 6 months ago

Not necessarily, but feel free to experiment

alonsojr1980 commented 6 months ago

FYI, lightning LORA (sdxl_lightning_8step_lora.safetensors) makes it burn even with low scale. Other than that, it's working great. Thanks!

pamparamm commented 6 months ago

Thanks for the issue BTW. It's actually a bug - applying PAG before and after LoRAs should produce same pictures. I'll try to fix that