[Bug] Result is Black Image when using Flux Kontext with fp4 on Blackwell (5090)

[ech3lon42]

Checklist

• 1. I have searched for related issues and FAQs (https://github.com/mit-han-lab/nunchaku/blob/main/docs/faq.md) but was unable to find a solution.
• 2. The issue persists in the latest version.
• 3. Please note that without environment information and a minimal reproducible example, it will be difficult for us to reproduce and address the issue, which may delay our response.
• 4. If your report is a question rather than a bug, please submit it as a discussion at https://github.com/mit-han-lab/nunchaku/discussions/new/choose. Otherwise, this issue will be closed.
• 5. If this is related to ComfyUI, please report it at https://github.com/mit-han-lab/ComfyUI-nunchaku/issues.
• 6. I will do my best to describe the issue in English.

Describe the Bug

Running the flux kontext workflow (nunchaku-flux.1-kontext-dev.json) using fp4 model on 5090 (svdq-fp4_r32-flux.1-kontext-dev) results in black image.

Using

• updated ComfyUI
• nunchaku wheel 0.3.1 for torch 2.7 (installed using wheel download workflow)
• nunchaku @ https://huggingface.co/mit-han-lab/nunchaku/resolve/main/nunchaku-0.3.1+torch2.7-cp311-cp311-win_amd64.whl#sha256=6ccd7600d1ad6f9bb93fc624a14b9769ce67b94a597c958380d92ef302538b42
• ComfyUI-nunchaku version: 0.3.3
• torch==2.7.0+cu128
• sageattention @ https://github.com/woct0rdho/SageAttention/releases/download/v2.1.1-windows/sageattention-2.1.1+cu128torch2.7.0-cp311-cp311-win_amd64.whl#sha256=629da9f19e810f98c8eb1d4d9af07cd29b500069b358689ff9bc48defe057795
• 5090, latest drivers
• CUDA 12.8

Nunchaku with svdq-fp4-flux.1-dev works normally. Flux Kontext (without nunchaku) also works normally.

Full Log log_nunchaku031_ComfyUINunchaku033_black_image_5090_fp4.txt

Am I doing something wrong or is this a bug?

Environment

Windows, 5090, CUDA 12.8, Torch 2.7.0+cu128, Sage Attention 2.1, Nunchaku Wheel 0.3.1, NunchakuComfyUI 0.3.3

Full environment: pip_freeze.txt

Reproduction Steps

Running the flux kontext workflow (nunchaku-flux.1-kontext-dev.json) using svdq-fp4_r32-flux.1-kontext-dev results in black image on 5090.

[lmxyy]

Could you try setting the cache_threshold threshold to 0 and try this example image: https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/yarn-art-pikachu.png

[lmxyy]

Duplicated. Merged into nunchaku-ai/ComfyUI-nunchaku#336

[ech3lon42]

I set up a completely new ComfyUI instance and only installed nunchaku. Still didn't work.

Then I decided to delete the model and re-download it from huggingface:
https://huggingface.co/mit-han-lab/nunchaku-flux.1-kontext-dev/blob/main/svdq-fp4_r32-flux.1-kontext-dev.safetensors

And now it works!!

So it seems I had a broken or incomplete model file. Unfortunately this was not transparent to me since I had no errors, just a black image. So everything is fine with the model on 5090... sorry for the confusion. If anyone runs into this black image problem - ensure the integrity of your model files!