ostris
When using Z-Image LoRA with ComfyUI, the message "lora key not loaded" appears.
Do I need to wait for ComfyUI support or do I need some kind of converter? (LoRA clearly works)
lora key not loaded: diffusion_model.layers.9.attention.to_out.0.lora_A.weight lora key not loaded: diffusion_model.layers.9.attention.to_out.0.lora_B.weight lora key not loaded: diffusion_model.layers.9.attention.to_q.lora_A.weight lora key not loaded: diffusion_model.layers.9.attention.to_q.lora_B.weight lora key not loaded: diffusion_model.layers.9.attention.to_v.lora_A.weight lora key not loaded: diffusion_model.layers.9.attention.to_v.lora_B.weight
Use this lora loader
https://github.com/PGCRT/CRT-Nodes
https://www.youtube.com/watch?v=Dkffab2QBJM
ComfyUI added support for it (I suppose), still pending but it will soon be available for update https://github.com/comfyanonymous/ComfyUI/commit/52a32e2b323b90295ab05a8c299590c890f2ecb6
This is fixed, but one thing I noticed is that the quality of the LoRA during training/sampling is noticeably better than when used in Comfy. I had the same issue with SDXL and Chroma, too.
This is fixed, but one thing I noticed is that the quality of the LoRA during training/sampling is noticeably better than when used in Comfy. I had the same issue with SDXL and Chroma, too.
Interestingly I can confirm this is the case on my side as well. I have been trying to recreate the excellent quality output I had during training/sampling with Ostris since yesterday, but I cannot achieve the same quality no matter what I try in Comfy.
