Updated: Mar 25, 2026
base model
Nobody has yet uploaded this amazing Nunchaku version of Qwen Image 2512 to Civitai so I am doing it.
It can make amazing 2K images in under 60 seconds that take 300 Seconds+ on the normal Qwen 2512 version.
It is Lora compatible with the right custom nodes.
Nunchaku Qwen Image LoRA Stack V3: https://github.com/ussoewwin/ComfyUI-QwenImageLoraLoader
I have been using this V3 Wuli 4 step Lightning lora: https://huggingface.co/Wuli-art/Qwen-Image-2512-Turbo-LoRA/blob/main/Wuli-Qwen-Image-2512-Turbo-LoRA-4steps-V3.0-bf16.safetensors (it is very good)
Original source https://huggingface.co/QuantFunc/Nunchaku-Qwen-Image-2512
Optional:
If you want to use the ClownShark sampler (which is slightly better in my testing) You have to apply this Pull request: https://github.com/nunchaku-ai/ComfyUI-nunchaku/archive/refs/pull/600/head.zip
to \custom_nodes\ComfyUI-nunchaku\
to fix an error with the node and Nunchaku combination:
Expected all tensors to be on the same device, but got mat1 is on cpu, different from other tensors on