QuantStack/Wan2.1_14B_VACE-GGUF · How to load VACE-GGUF and CausVid lora in diffusers ?

I'm attempting to re-implement the ComfyUI VACE workflow using Diffusers, but I encountered an error when trying to load the GGUF-quantized transformer model. Is there a supported way to load GGUF-quantized models in Diffusers for VACE?

Attempted Approach

I tried loading the model using WanVACETransformer3DModel.from_single_file() with the following code:

transformer = WanVACETransformer3DModel.from_single_file(
    ckpt_path,
    config="vace/transformer/config.json",
    quantization_config=GGUFQuantizationConfig(compute_dtype=torch.bfloat16),
    torch_dtype=torch.bfloat16,
)

Error

The code fails with:

ValueError: FromOriginalModelMixin is currently only compatible with [StableCascadeUNet, UNet2DConditionModel, ...], but not WanVACETransformer3DModel.

Additional Context

GGUF Model: QuantStack/Wan2.1_14B_VACE-GGUF
Workflow Reference: VACE Example Workflow

Any guidance or alternative approaches would be greatly appreciated!