Fix: allow Flux Fill pipeline to skip image preprocessing when latents are provided

[Xingfu-Yi]

Fix: allow Flux Fill pipeline to skip image preprocessing when latents are provided

What does this PR do?

Fixes a critical bug in the Flux Fill pipeline where image was implicitly assumed to exist even when latents and masked_image_latents were directly provided as inputs.

When both latents and masked_image_latents are passed, supplying an image is not only unnecessary but also disallowed by input validation. However, the current implementation still tries to preprocess image, which results in a NoneType error when image is None.

This PR adds a minimal and safe fix:
If image is None, the code now explicitly sets init_image = None and skips unnecessary preprocessing. This ensures that downstream logic, such as self.prepare_latents(), correctly bypasses latent initialization and uses the user-provided latents as intended.

Changes

• Added a conditional check to skip image preprocessing when image is None
• Prevents runtime errors when using latent-only initialization
• Maintains compatibility and minimal code change within the existing pipeline logic

Why is this important?

Without this fix, initializing the Flux Fill pipeline with precomputed latents is impossible due to a NoneType dereference during image preprocessing.
This breaks valid workflows that rely on directly supplying latents and masked_image_latents (for example, when reusing intermediate latent states for iterative or compositional generation).

Fixes

Fixes # (issue number if applicable)

Before submitting

• This PR fixes a runtime bug (not a docs or typo fix).
• I have read the [contributor guideline](https://github.com/huggingface/diffusers/blob/main/CONTRIBUTING.md).
• I have added or verified tests for the affected path.
• I have ensured no behavioral changes when image is provided normally.