[Bug] The LORA model converted from FAL.AI does not support

[T8mars]

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/ComfyUI-nunchaku/discussions/new/choose. Otherwise, this issue will be closed.
• 5. I will do my best to describe the issue in English.

Describe the Bug

This error message will be displayed, while other models are functioning normally. Many FAL.AI trained models have this issue Reproducible error LORA download address Link: https://pan.quark.cn/s/7a5d0ace4d32

Environment

python310 cuda128 pytorch2.7

Reproduction Steps

This error message will be displayed, while other models are functioning normally. Many FAL.AI trained models have this issue Reproducible error LORA download address Link: https://pan.quark.cn/s/7a5d0ace4d32

[JackDainzh]

Can confirm the same bug happening with the Kontext lora I've made with comfy convertion selected on Fal.ai.

[JackDainzh]

https://pastebin.com/fxq4mWHe

Thanks to @akedia for providing the initial code snippet so I could bounce off of it. This script converts ONLY the final layers of lora made and converted to comfy from Fal.ai (meaning when creating the lora, you should've selected the comfy backend, not fal, although it is possible to convert for fal, I can just adjust the code accordingly if it is needed).

[StartHua]

同样遇到这个问题，有人解决了？

[Bluear7878]

同样遇到这个问题，有人解决了？

I’m currently investigating this issue and working on a fix.

[JackDainzh]

同样遇到这个问题，有人解决了？

I’m currently investigating this issue and working on a fix.

Isn't it just a matter of assigning missing keys as I've suggested?

[Bluear7878]

@T8mars
Can you share your workflow script?

[RamonGuthrie]

My LoRA model converter fixes issues with LoRA models (Flux Dev and Flux Dev Kontext LoRAs) trained and produced with FAL.AI, which currently do not work correctly with ComfyUI-Nunchaku nodes. The converted LoRAs also work in regular LoRA loader nodes!

Github: 🔗 https://github.com/RamonGuthrie/ComfyUI-RBG-LoraConverter

[Bluear7878]

@JackDainzh
Thank you for pointing this out. The root cause was indeed the discrepancy between Nunchaku’s and FAL.AI’s LORA key naming conventions. I’ve addressed this in the PR (nunchaku-ai/nunchaku#557). Please let me know if you encounter any further issues.