2509 loras barely make any difference?
#13
by newgenerator - opened
I've downloaded and tested various 2509 loras at different strengths with both R32 and R128 FP4 variants, they barely make any difference to the image. Can someone tell me if I'm doing something wrong, I'm using the comfyui default "load lora" node.
I don't think LoRAs work with Nunchaku Qwen or Qwen Edit yet.
Correction 27.10.2025: i found this node for use lora with Nunchaku SVDQuant!!
https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader
Just follow the installation guide and work for me ! Hope this can help.
OLD: Loras dosen't work on Nunchaku qwen and qwen-edit. For use loras, you must use normal models or gguf from quantstack. I test and work on normal and gguf.