download_from_original_stable_diffusion_ckpt does not load SDXL checkpoints without specifying pipeline_class

[allenbenz]

Describe the bug

94b332c changed a hardcoded class reference of StableDiffusionXLPipeline to pipeline_class, pipeline_class defaults to StableDiffusionPipeline so it tries to load the SDXL checkpoint as a SD checkpoint.

Providing the argument pipeline_class=StableDiffusionXLPipeline to the download_from_original_stable_diffusion_ckpt works around the issue.

Reproduction

from diffusers.pipelines.stable_diffusion.convert_from_ckpt import download_from_original_stable_diffusion_ckpt

download_from_original_stable_diffusion_ckpt(
    checkpoint_path="E:/models/sd_xl_base_1.0_0.9vae.safetensors",
    original_config_file="E:/models/sd_xl_base_1.0_0.9vae.yaml",
    from_safetensors=True
)

Logs

Traceback (most recent call last):
  File "E:\tmp\repro.py", line 3, in <module>
    download_from_original_stable_diffusion_ckpt(
  File "E:\tmp\venv\lib\site-packages\diffusers\pipelines\stable_diffusion\convert_from_ckpt.py", line 1592, in download_from_original_stable_diffusion_ckpt
    pipe = pipeline_class(
TypeError: StableDiffusionPipeline.__init__() got an unexpected keyword argument 'text_encoder_2'

System Info

• diffusers version: 0.20.0.dev0
• Platform: Windows-10-10.0.19045-SP0
• Python version: 3.10.7
• PyTorch version (GPU?): 2.0.1+cu118 (True)
• Huggingface_hub version: 0.16.4
• Transformers version: 4.31.0
• Accelerate version: 0.21.0
• xFormers version: 0.0.20
• Using GPU in script?: yes
• Using distributed or parallel set-up in script?: no

Who can help?

@yiyixuxu

[patrickvonplaten]

cc @yiyixuxu can you take a look?

[yiyixuxu]

@allenbenz
this should fix it
#4461

[sayakpaul]

Closing since that has been fixed.