OpenMOSS / OpenMOSS/MOVA

High failure rate in T2AV generation

Open
#38 1 comment 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
1.1k
Forks
93
Avg merge
1m
Merged PRs (30d)
1

Description

Describe
When using the MOVA pipeline for pure T2V/T2AV generation, I am experiencing a highly unstable generation process. Approximately 40% of the generated videos result in corrupted, solid-color outputs (the entire video is just a flat color with no coherent structures or details).

Image ![Image](https://github.com/user-attachments/assets/7461b429-a023-47cb-9b39-52f7db79d3d8)

Following the standard T2V approach for MOVA, I am passing a pure white PIL.Image as the image condition to the pipeline. I strongly suspect the issue lies in how the pipeline_mova.py loads, encodes, or concatenates this pure white frame in the prepare_latents stage, leading to a latent collapse during the diffusion process.

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

Start in pipeline_mova.py, focusing on the prepare_latents stage where the pure white PIL.Image is loaded, encoded, and concatenated for standard T2V generation. Reproduce the MOVA T2V/T2AV flow and compare successful videos with the corrupted solid-color outputs. Done means the generation process no longer produces the reported latent-collapse failures at the stated rate.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
audio-video-rtc, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.