Difference Between flux-controlnet-canny-v3 and flux-controlnet-canny-diffusers

#1
by yzqy66 - opened

Hi everyone,

Can someone explain the main differences between the XLabs-AI/flux-controlnet-canny-v3 and XLabs-AI/flux-controlnet-canny-diffusers models? I'm curious about their features and use cases.

Thanks!

XLabs AI org

Hi! Thank you for your question. This version includes converted weights from the XLabs-AI/flux-controlnet-canny-v3 model, so it can be used in the Diffusers library. We prepared a PR with this feature: https://github.com/huggingface/diffusers/pull/9638.

In terms of quality and training details, it's the same model version, the ControlNet canny-v3.

Hi @anghellia ,

I tried flux-canny-controlnet-v3.safetensors with stable-diffusion.cpp, but got an error tensor 'input_hint_block.10.bias' has wrong shape in model file: got [16, 1, 1, 1], expected [96, 1, 1, 1]. I filed an issue on stable-diffusion.cpp. I just wonder if flux-controlnet-canny-v3 and flux-controlnet-canny-diffusers can run on stable-diffusion.cpp? Thanks a lot!

XLabs AI org

Hi @apepkuss79 ,
I checked your log on GitHub. It seems the model isn't loading at all, as it's throwing a warning: 'unknown tensor' for the ControlNet weights.

The models flux-controlnet-canny-v3 and flux-controlnet-canny-diffusers have specific layers in their architecture that aren't supported in stable-diffusion.cpp, so it's not possible to run them.

@anghellia Thank you for the quick reply. I'll post a feature request on stable-diffusion.cpp for supporting these models. Thank you so much!

Sign up or log in to comment