Fix get_dummy_inputs for Stable Diffusion Inpaint Tests

[dg845]

What does this PR do?

This PR fixes the get_dummy_inputs method for StableDiffusionInpaintPipelineFastTests. Currently, get_dummy_inputs creates a black image and black mask_image, which means that we are technically not repainting the image at all, which doesn't seem like the right behavior.

This PR is a spin-off of #4536, where the above issue was noticed (see #4536 (comment)).

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline?
• Did you read our philosophy doc (important for complex PRs)?
• Was this discussed/approved via a Github issue or the forum? Please add a link to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

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
@yiyixuxu

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[patrickvonplaten]

Hey @dg845, thanks a lot for the PR - the change makes a lot of sense!

Could you maybe try to update the failing tests? Think then we can merge

[dg845]

The tests which are currently failing are:

• test_stable_diffusion_inpaint.py::StableDiffusionInpaintPipelineFastTests::test_stable_diffusion_inpaint
• test_stable_diffusion_inpaint.py::StableDiffusionSimpleInpaintPipelineFastTests::test_stable_diffusion_inpaint

which makes sense, given that the input image and mask have changed. @patrickvonplaten, do you know what reference implementation(s) were used to get the expected slices for StableDiffusionInpaintPipeline testing? Based on the slow tests using the runwayml/stable-diffusion-inpainting checkpoint for the dedicated inpainting model and runwayml/stable-diffusion-v1-5 for the simple text-to-image model, I assume the reference implementation for both was runwayml/stable-diffusion.

(Looking through the code, the only fast tests that use an expected_slice are the two tests listed above. I guess we aren't forced to use the same reference implementation to get the new expected_slices, but it seems natural to me to try to keep it consistent with the previous expected_slices, and with the reference implementation used to get the expected_slices for the slow tests.)

[patrickvonplaten]

Hey @dg845,

Could you maybe just update the expected_slice to match the results you're getting in this PR? Those fast test are "dummy" integration tests

[dg845]

I have updated the expected_slices for the *.test_stable_diffusion_inpaint tests to match the pipeline outputs (e.g. the output of sd_pipe(**inputs).images).

[patrickvonplaten]

Test failure is unrelated - merging!