Here are the Nunchaku 4 & 8 Step Lightning models for Improved Qwen Image Edit v2509
Here is official information: nunchaku-tech/nunchaku-qwen-image-edit-2509 · Hugging Face
Description
FAQ
Comments (2)
This model cause ComfyUI crash without error message.
Need workflow or at least description of correct node for loading it.
I downloaded the 8 step version, I have 2 cards RTX3090 & RTX4080 , I used a Nunchaku Qwen DiT loader node to load this model, I got this ERROR:
Please use "int4" quantization for Turing, Ampere and Ada GPUs.
& I tried to load it through a (Load Checkpoint) node & the Sampler gives me colorful noisy image (colorful snowflakes) ONLY !
Be careful & you better Don't download this , first See What is your GFX card: RTX 5000 series need fp4, RTX 3000/4000 need int4.