Describe the bug

Flux models are sometimes distributed as transformer only, because all the other pipeline components are usually not changed.

Examples include:

• finetunes on CivitAI, to offer a smaller download size

• https://huggingface.co/lllyasviel/flux1-dev-bnb-nf4/tree/main

Since support for Flux Fill was added here 862a7d5, diffusers.FluxTransformer2DModel.from_single_file fails to load these files:

File "/.../src/diffusers/src/diffusers/loaders/single_file_utils.py", line 598, in infer_diffusers_model_type
    if checkpoint["img_in.weight"].shape[1] == 384:
       ~~~~~~~~~~^^^^^^^^^^^^^^^^^
KeyError: 'img_in.weight'

Reproduction

Download one of the example mentioned above, and run

from diffusers import FluxTransformer2DModel

transformer = FluxTransformer2DModel.from_single_file(your_file)

Logs

No response

System Info

HEAD

Who can help?

@DN6