[LCM] Make sure img2img works

[patrickvonplaten]

What does this PR do?

This PR cleans the LCM pipeline and scheduler and also adds LCM Img2Img to "main" as well as the auto pipeline.

The following now works:

from diffusers import AutoPipelineForText2Image, AutoPipelineForImage2Image
import torch

pipe = AutoPipelineForText2Image.from_pretrained("SimianLuo/LCM_Dreamshaper_v7", revision="c4c79506b49efd22e0cd254ec4092ab3d7d8ae52", local_files_only=True)

prompt = "A beautiful landscape of a snowy mountain"

num_inference_steps = 2

image = pipe(prompt=prompt, num_inference_steps=num_inference_steps, output_type="pil").images[0]
image.save("snow_mountain.png")

pipe = AutoPipelineForImage2Image.from_pipe(pipe)

prompt = "A beautiful landscape of a very red mountain"

image = pipe(prompt=prompt, image=image, num_inference_steps=4, output_type="pil").images[0]
image.show()

cc @vladmandic

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[vladmandic]

thanks for really quick turnaround.
btw, from_single_file for LCM is not implemented?

ValueError: Unhandled pipeline class: LatentConsistencyModelPipeline

[sayakpaul]

btw, from_single_file for LCM is not implemented?

Not yet.

[jon-chuang]

What is the reason for this special case?