aigc-apps / aigc-apps/VideoX-Fun

FUN-Control models have poor quality with zero conditioning

Open
#451 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
2.3k
Forks
188
Avg merge
2d 22h
Merged PRs (30d)
3

Description

Thank you for your amazing work!

I noticed that when using the Wan FUN-Control models without control video (i.e., setting control latents to zero tensor), the generation quality is very poor. Even worse than the base Wan model. Here are examples with Wan2.1 1.3b. Top is FUN-Control, bottom is base Wan.

Do you have any insights on why this is the case? When using sparse control videos like openpose, the uncontrolled parts of the video have similar poor quality.

https://github.com/user-attachments/assets/fb11e678-16e2-4e60-bc49-4c5d9b00dd58

https://github.com/user-attachments/assets/52b68cc6-0610-4dfc-b6bd-d7e0007d006c

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No source files or tests are named. Start by reproducing the Wan2.1 1.3b comparison with zero control latents and sparse OpenPose control, then trace the FUN-Control conditioning path and compare it with the base Wan path. Done means the uncontrolled output no longer shows the reported quality degradation.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.