Fix self.vae reference in modular SDXL inpaint VAE encoder

[Devam0311]

Issue link: this addresses Issue 5 of #13610.
I have deliberately not used a Fixes keyword: #13610 tracks 7 separate findings, so a closing keyword would close it as soon as this one merges while the others are still open. Happy for the no-issue-needed label to be applied, or I can add a closing keyword if you would rather the review issue be closed and reopened.

What does this PR do?

Fixes Issue 5 of #13610 (stable_diffusion_xl model/pipeline review).

StableDiffusionXLInpaintVaeEncoderStep._encode_vae_image reads self.vae.config.scaling_factor, but self here is the block, not the components object:

if latents_mean is not None and latents_std is not None:
    ...
    image_latents = (image_latents - latents_mean) * self.vae.config.scaling_factor / latents_std
    #                                                ^^^^ block, not components
else:
    image_latents = components.vae.config.scaling_factor * image_latents   # correct, two lines below

Reproduced on main:

AttributeError: 'StableDiffusionXLInpaintVaeEncoderStep' object has no attribute 'vae'

So any inpaint modular pipeline whose VAE config defines latents_mean and latents_std fails outright. The equivalent non-inpaint encoder in the same file already uses components.vae.

Solution

One-word fix — self.vae → components.vae — matching the else branch directly below it and the non-inpaint encoder above it.

Testing

TestSDXLInpaintVaeEncoderStep::test_encode_vae_image_with_latents_mean_and_std drives the method with a stand-in VAE, as the issue's reproduction does. That is deliberate rather than lazy: the branch only executes for a VAE config carrying latents_mean / latents_std, which the SDXL VAE used elsewhere in that test file does not define, so the existing fixtures cannot reach it.

With latents_mean=0 and latents_std=1 the branch reduces to a plain scaling_factor multiply, so the test asserts the exact expected values rather than just the absence of an exception.

Verified it catches the bug: with the fix reverted it fails with the AttributeError above.

pytest tests/modular_pipelines/stable_diffusion_xl/   -> 47 passed, 16 skipped
ruff check / ruff format --check                      -> clean
git diff --check                                      -> clean

Before submitting

• Did you use an AI agent (Claude Code, Codex, Cursor, etc.) to help with this PR? If so:
  • Did you read the Coding with AI agents guide?
  • Did you run the self-review skill on the diff?
  • Did you share the final self-review notes in the PR description or a comment?
• Did you read the contributor guideline?
• Was this discussed/approved via a GitHub issue? — stable_diffusion_xl model/pipeline review #13610
• Did you write any new necessary tests?

Self-review notes

Reviewed against .ai/references/review-rules.md, modular.md and testing.md. No blocking issues.

For the reviewer:

1. The new test class subclasses nothing and holds no pipeline fixtures, unlike the rest of that file. That is intentional — it is a unit test of one method on one block, and building a full modular pipeline with a latents_mean/latents_std VAE just to reach one line seemed disproportionate. Happy to fold it into the existing tester config instead if you'd prefer consistency.
2. This is a pure bug fix: the branch could not previously execute without raising, so there is no behaviour to preserve.
3. Scope kept to Issue 5. Issues 2 and 4 of stable_diffusion_xl model/pipeline review #13610 are in Use negative crop coordinates when building SDXL negative time ids #14676 and Do not watermark or postprocess SDXL latent output #14678.

Who can review?

@yiyixuxu @asomoza @DN6

[github-actions]

Hi @Devam0311, thanks for the PR! It does not appear to link an issue it fixes. If this PR addresses an existing issue, please add a closing keyword (e.g. Fixes #1234) to the PR description so the issue is linked. See the contribution guide for more details. If this PR intentionally does not fix a tracked issue, a maintainer can add the no-issue-needed label to silence this reminder.

Please note that PRs without a linked issue are likely to be automatically closed 10 days after this notice.

Once the PR links an issue (or gets the no-issue-needed label), you can ignore this message — it stays here as a comment, but it no longer applies.

[adi-IL]

Tested locally on Linux x86_64 with python 3.14 and torch 2.13.

Confirmed that StableDiffusionXLInpaintVaeEncoderStep._encode_vae_image raised AttributeError: 'StableDiffusionXLInpaintVaeEncoderStep' object has no attribute 'vae' on main whenever latents_mean and latents_std were present, because vae lives on components, not self.

With this fix to components.vae.config.scaling_factor, the normalization branch runs cleanly and matches the fallback in the else branch. The new unit test passes.