[dhvani7014]

What does this PR do?

Fixes #15071

Deprecates the output_type == "latent" branches in the modular decode blocks for flux, sd3, sdxl, cosmos and echo, targeting removal in 0.47.0 alongside the packed-latents deprecation from #14737. Latents are available via output="latents" instead. ltx2 is left for #14612.

Self-review notes

• Passing output_type="latent" to these decoders still returns the latents as before, but now emits a FutureWarning (via deprecate, 0.47.0) pointing to output="latents".
• Any other output_type not in ("pil", "np", "pt") raises a ValueError, same as minimax_h3.
• Rebased onto main after move unpack out of the flux, krea2 and ltx decoders into the core denoise groups #14737 merged and kept its packed-latents deprecation in the flux decoder.
• Tests that used "latent" now read latents/audio_latents from the output or use output_type="pt". Added a flux test that checks the deprecation warning and that the latents come back unchanged.
• Updated the echo docs line that described the latent output.
• Ran make style and pytest on tests/modular_pipelines/{flux,echo,stable_diffusion_3,stable_diffusion_xl}: 211 passed, 28 skipped. The 2 SDXL test_to_device failures also fail on main on my Mac (MPS), so they're unrelated.

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?
• 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. Follow-up to #14737: drop output_type="latent" from modular decode blocks #15071
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?
• Are you the author (or part of the team) of the model/pipeline (only applicable for model/pipeline related PRs)?

Who can review?

@yiyixuxu

[dhvani7014]

Hi @yiyixuxu, could you please review this PR whenever you get a chance? I'd really appreciate your feedback. Thank you!

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[yiyixuxu]

thanks!
i left a comment

[dhvani7014]

@yiyixuxu Done, switched to a deprecation targeting 0.47.0, matching #14737. Also rebased onto main and added a test for the warning

[dhvani7014]

Hi @yiyixuxu, just a gentle ping on this one when you have time. I switched to the deprecation as you suggested. Let me know if anything else needs changing!

[dhvani7014]

Thanks @yiyixuxu! Updated the deprecation message, removed the extra output_type check (kept just the deprecate in all the decoders), and added a line in echo.md on getting the denormalized VAE tensors without decoding.

[yiyixuxu]

Suggested change

	without decoding, run the pipeline without the decode step and denormalize these with `latents_mean`, `latents_std`
	and `vae_scaling_factor` for video, or `audio_latents_mean` and `audio_latents_std` for audio.
	without decoding, run the pipeline without the decode step.