LoRA weights of Stable Diffusion XL for fast image generation.
Important Usage Guidance
Use DDIM or Euler instead of LCM for sampling!
The name of each LoRA weights indicates how many inference steps they should be applied.
The name of each LoRA weights indicates whether they are able to use normal CFGs or small CFGs.
NormalCFG means that model equipped with the LoRA can use CFG value 2-9 for generation. For example, Juggernaut XL, in my test, works with 3-5. Hopefully I just find a better way to train normalCFG LoRAs with better visual quality, will update them in the coming days!!!
SmallCFG means that the model equipped with the LoRA can use CFG value 1-2 for generation.
About the performance of normal CFG LoRAs.
Important Information
[[paper](https://arxiv.org/abs/2405.18407)]
[[code](https://github.com/G-U-N/Phased-Consistency-Model)]
[[huggingface](https://huggingface.co/wangfuyun)]
Description
FAQ
Comments (7)
Very interest in trying this! When I try in Comfy though the LoRA is coming up w/ lora key not loaded. It should work like a normal LoRA, right? Can you maybe post a step by step for Comfy or A1111?
You can get the converted weights here: https://huggingface.co/Kijai/converted_pcm_loras_fp16/tree/main
Thank you.
Looks like the whole set of CFG/step combinations got dumped into one huge ZIP file while they are individually downloadable on HF.
Any chance we can get them uploaded individually here as well?
LIST OF ALL 7 FILES inside:
1 - pcm_sdxl_normalcfg_4step
2 - pcm_sdxl_normalcfg_8step
3 - pcm_sdxl_normalcfg_16step
4 - pcm_sdxl_smallcfg_2step (PREFERRED in comfyui for those who want ULTRA speed during txt2img , then refine or UPSAMPLE it later with normal 20+ STEPS)
5 - pcm_sdxl_smallcfg_4step
6 - pcm_sdxl_smallcfg_8step
7 - pcm_sdxl_smallcfg_16step
You're welcome!
"4 - pcm_sdxl_smallcfg_2step (PREFERR....."
Wow! You are very right, I tested it and it seems to work, nice tip!😄
How is it compared to HyperSD?