diffusers
[SDXL] Improve docs
#4196
Merged

[SDXL] Improve docs #4196

patrickvonplaten merged 7 commits into main from improve_sd_xl_docs_again
patrickvonplaten
patrickvonplaten2 years ago (edited 2 years ago)

What does this PR do?

Adds missing docs for micro conditioning and corrects inpainting example for SD-XL

Before submitting

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

patrickvonplaten Improve docs
73548fcd
stevhliu
stevhliu commented on 2023-07-21
src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py
670670 [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf).
671671 Guidance rescale factor should fix overexposure when using zero terminal SNR.
672672 original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
673 TODO
673
Part of SDXL's micro-conditioning as explained in section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
674
If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled. `original_size` defaults to `(width, height)` if not specified.
stevhliu2 years ago

What do you think about shortening it to like the below so it's a bit shorter/easier to read (big ➕ for linking to hf.co/papers!), and I think it'd be clearer to lead with what the parameter does?

Suggested change
Part of SDXL's micro-conditioning as explained in section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled. `original_size` defaults to `(width, height)` if not specified.
If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled. `original_size` defaults to `(width, height)` if not specified (see [SDXL micro-conditioning](https://huggingface.co/papers/2307.01952) for more details).
patrickvonplaten2 years ago

Swapped the two sentences - would like to keep the "section 2.2" part so that people don't have to search the whole paper. Let me know what you think :-)

stevhliu2 years ago

Nice, that works for me! 👍

patrickvonplaten Correct docs
a9463924
patrickvonplaten Add better example inpaint
d726b39d
patrickvonplaten make style
0f1b2d34
patrickvonplaten Merge branch 'main' into improve_sd_xl_docs_again
c3b41122
HuggingFaceDocBuilderDev
HuggingFaceDocBuilderDev2 years ago (edited 2 years ago)

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

patrickvonplaten patrickvonplaten requested a review from stevhliu stevhliu 2 years ago
stevhliu
stevhliu approved these changes on 2023-07-24
stevhliu2 years ago

Awesome! 👍

Conversation is marked as resolved
Show resolved
src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py
791 not specified it will default to `(width, height)`. Part of SDXL's micro-conditioning as explained in
792 section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
785793 aesthetic_score (`float`, *optional*, defaults to 6.0):
786 TODO
794
Part of SDXL's micro-conditioning as explained in section 2.2 of
795
[https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
796
simulate an aesthetic score of the generated image by influencing the positive text condition.
stevhliu2 years ago

Let's also switch the sentence order here and for negative_aesthetic_score as well :)

Suggested change
Part of SDXL's micro-conditioning as explained in section 2.2 of
[https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
simulate an aesthetic score of the generated image by influencing the positive text condition.
Used to simulate an aesthetic score of the generated image by influencing the positive text condition. Part of SDXL's micro-conditioning as explained in section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
Conversation is marked as resolved
Show resolved
src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py
990 not specified it will default to `(width, height)`. Part of SDXL's micro-conditioning as explained in
991 section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
992 aesthetic_score (`float`, *optional*, defaults to 6.0):
993
Part of SDXL's micro-conditioning as explained in section 2.2 of
994
[https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
995
simulate an aesthetic score of the generated image by influencing the positive text condition.
stevhliu2 years ago

Same here and for negative_aesthetic_score:

Suggested change
Part of SDXL's micro-conditioning as explained in section 2.2 of
[https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
simulate an aesthetic score of the generated image by influencing the positive text condition.
Used to simulate an aesthetic score of the generated image by influencing the positive text condition. Part of SDXL's micro-conditioning as explained in section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
patrickvonplaten Apply suggestions from code review
b950e073
patrickvonplaten fix
1d79750d
patrickvonplaten patrickvonplaten merged d74561da into main 2 years ago
kashif kashif deleted the improve_sd_xl_docs_again branch 1 year ago

Login to write a write a comment.

Login via GitHub

Reviewers
Assignees
No one assigned
Labels
Milestone