wan: vae: Fix light/color issue

[rattus128]

#13090

There was an issue where the resample split was too early and dropped one of the rolling convolutions a frame early. This is most noticable as a lighting/color change between pixel frames 5->6 (latent 2->3), or as a lighting change between the first and last frame in an FLF wan flow.

Example test case:

WAN2.2 FLF 640x640x81f
Same image F&L

Before:

Screencast.from.2026-03-22.08-25-24.webm

After:

Screencast.from.2026-03-22.08-30-19.webm

VRAM consumption checked as still lower than 17.x ✅

Regression Tests:
WAN 2.2 I2V ✅
WAN 2.2 T2V ✅

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro

Run ID: 7607fef4-7795-43b9-a95b-a10c8b20de4a

📥 Commits

Reviewing files that changed from the base of the PR and between 11c15d8 and 182e0ba.

📒 Files selected for processing (1)

• comfy/ldm/wan/vae.py

---

📝 Walkthrough

Walkthrough

The Decoder3d.run_up method in the VAE module was refactored to change the handling of Resample layers with mode == 'upsample3d'. The layer is now applied to the input before branching on the upsample3d condition. The threshold for triggering upsample3d processing changed from x.shape[2] > 1 to x.shape[2] > 2, and the slicing strategy shifted from per-frame chunks of length 1 to chunks of length 2. Additionally, the recursive call now increments layer_idx instead of preserving it.

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title 'wan: vae: Fix light/color issue' is directly related to the changeset, which fixes a lighting/color issue in the WAN VAE decoder by adjusting the resample split timing.
Description check	✅ Passed	The description clearly explains the issue (resample split too early, dropping rolling convolutions prematurely) and provides concrete test cases with before/after evidence and regression test results.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}