updated doc for stable diffusion pipelines

[yiyixuxu]

• one doc page for each stable diffusion pipeline
• added docstring examples (text2img, inpaint, img2img, upscale, depth2img)

[HuggingFaceDocBuilderDev]

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

[yiyixuxu]

Hi @anton-l :

How can I fix these 2 tests that failed?
I don't understand what the error I got from check_code_quality test means
for the check_repository_consistency - do I need to copy and modify the same docstring example from stable diffusion pipeline to alt diffusion? or is this something I can ignore?

[anton-l]

@yiyixuxu most of the time the code quality and repo consistency tests can be fixed by running make style and make fix-copies from the repo's root. The dependencies for those scripts can be installed with pip install diffusers[quality] ✨
The definitions are gathered here, if you ever need to run the steps manually: https://github.com/huggingface/diffusers/blob/main/Makefile#LL57C3-L57C3

There's also make quality that will check for things that need manual attention (the check_code_quality test does the same thing)

[patrickvonplaten]

nice!

[patrickvonplaten]

Could we maybe add new sections for every example here?:

Text-to-Image

Text-to-Image (512x512 resolution): stabilityai/stable-diffusion-2-base with [StableDiffusionPipeline]
Text-to-Image (768x768 resolution): stabilityai/stable-diffusion-2 with [StableDiffusionPipeline]

```python
from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler
import torch
repo_id = "stabilityai/stable-diffusion-2-base"
pipe = DiffusionPipeline.from_pretrained(repo_id, torch_dtype=torch.float16, revision="fp16")
pipe.scheduler = DPMSolverMultistepScheduler.from_config(pipe.scheduler.config)
pipe = pipe.to("cuda")
prompt = "High quality photo of an astronaut riding a horse in space"
image = pipe(prompt, num_inference_steps=25).images[0]
image.save("astronaut.png")

Image Inpainting

etc...

This way we could better link to the different tasks.

[yiyixuxu]

for the upscaler should we use "Super-Resolution" or "Image-Upscaling"? I want to make the names consistent at least on the navigation bar

[patrickvonplaten]

"Super-Resolution" sounds good to me! Def in favor of making names consistent

[patrickvonplaten]

Looks nice - like the new structure much better & super cool that you added a bunch of examples to the pipeline_.....py functions :-)

Could we add a bit more text to the individual stable diffusion pipeline functions?

[yiyixuxu]

@yiyixuxu most of the time the code quality and repo consistency tests can be fixed by running make style and make fix-copies from the repo's root. The dependencies for those scripts can be installed with pip install diffusers[quality] ✨ The definitions are gathered here, if you ever need to run the steps manually: https://github.com/huggingface/diffusers/blob/main/Makefile#LL57C3-L57C3

There's also make quality that will check for things that need manual attention (the check_code_quality test does the same thing)

@anton-l when I run these commands I notice it changes files that I did not touch on this PR - why did this happen and is it ok for me to commit these changes?

[patrickvonplaten]

@yiyixuxu most of the time the code quality and repo consistency tests can be fixed by running make style and make fix-copies from the repo's root. The dependencies for those scripts can be installed with pip install diffusers[quality] sparkles The definitions are gathered here, if you ever need to run the steps manually: https://github.com/huggingface/diffusers/blob/main/Makefile#LL57C3-L57C3
There's also make quality that will check for things that need manual attention (the check_code_quality test does the same thing)

@anton-l when I run these commands I notice it changes files that I did not touch on this PR - why did this happen and is it ok for me to commit these changes?

It might be because of different versions of black, doc-builder and isort that you have installed.
For now, we pin black, doc-builder and isort to the versions as defined in setup.py:

black:

diffusers/setup.py

Line 83 in e4fe941

"black==22.8",

isort:

diffusers/setup.py

Line 91 in e4fe941

"isort>=5.5.4",

doc-builder:

diffusers/setup.py

Line 88 in e4fe941

"hf-doc-builder>=0.3.0",

So as you can see only black is actually pinned to 22.8, the others are not pinned.

Could you maybe do:

pip install black==22.8

=> then make style shouldn't change anything anymore.

@anton-l we should probably update black soon.

Also cc @patil-suraj @pcuenca @williamberman FYI

[yiyixuxu]

@yiyixuxu most of the time the code quality and repo consistency tests can be fixed by running make style and make fix-copies from the repo's root. The dependencies for those scripts can be installed with pip install diffusers[quality] sparkles The definitions are gathered here, if you ever need to run the steps manually: https://github.com/huggingface/diffusers/blob/main/Makefile#LL57C3-L57C3
There's also make quality that will check for things that need manual attention (the check_code_quality test does the same thing)

@anton-l when I run these commands I notice it changes files that I did not touch on this PR - why did this happen and is it ok for me to commit these changes?

It might be because of different versions of black, doc-builder and isort that you have installed. For now, we pin black, doc-builder and isort to the versions as defined in setup.py:

black:

diffusers/setup.py

Line 83 in e4fe941

"black==22.8",

isort:

diffusers/setup.py

Line 91 in e4fe941

"isort>=5.5.4",

doc-builder:

diffusers/setup.py

Line 88 in e4fe941

"hf-doc-builder>=0.3.0",

So as you can see only black is actually pinned to 22.8, the others are not pinned.

Could you maybe do:

pip install black==22.8

=> then make style shouldn't change anything anymore.

@anton-l we should probably update black soon.

Also cc @patil-suraj @pcuenca @williamberman FYI

Thanks! it works now

I added some descriptions for each pipeline - do you want to take a look and let me know if there is anything we need to add before I merge?

[patrickvonplaten]

Left some final suggestions to improve the docs, but overall this looks really nice to me :-)

Think once the suggestions are in we can merge this one!

Thanks a lot for working on it.

[yiyixuxu]

@anton-l @patrickvonplaten

test failing because the AltDiffusion pipelines have this "Copied from ..." statement but we want the doc string examples to be different: different checkpoint, prompt in a Chinese

this is one of the commits that triggered the error be35fda

Should I remove the "Copied from ..." statement?

[patrickvonplaten]

@yiyixuxu great catch! We'll do a trick here in this case to fix it - let me add one commit to your PR :-)

[patrickvonplaten]

@yiyixuxu this is more or less copied from transformers here: https://github.com/huggingface/transformers/blob/588faad1062198e45cf3aebed21dc1fc1e1ed0d7/src/transformers/utils/doc.py#L1130

[pcuenca]

Very nice!

[patrickvonplaten]

I've added a docstring function decorator we can use to circumvent this #Copied from" problem as it's very likely to happen again :-)

In short, for examples we should probably always recommend to:

• Define a global file variable called "EXAMPLE_DOCSTRING"
• Define an empty "Examples:" in the docstring
• Make use of the just added function decorator to replace it

This is a trick we also use in transformers to allow for both personalized example docstrings and be able to keep using the #Copied from mechanism