modelscope / modelscope/DiffSynth-Studio

[wan] [lora] train a longer wanvideo model

Open
#500 3 comments 1 reaction 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
13.1k
Forks
1.3k
Avg merge
13h 12m
Merged PRs (30d)
45

Description

Great thanks to the diffsynth work!
I found the base model Wan2.1_1.3B_t2v and Wan2.1_14B_t2v are designed to generate 81 frames of video. Even though I set the num_frames to 161, I found the generated video not so well(some flash shadow appeared and character move strangely). So I was wondering if someone has trained a longer version of t2v using diffsynth lora mode. Is there any mechanism in wan2.1 that causes it to only generate 81 frames of video? If I want to generate longer videos, is full training required, or can it be achieved through LoRA training?

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 files, tests, or implementation entry points are named. Start by reproducing the Wan2.1_1.3B_t2v and Wan2.1_14B_t2v behavior with num_frames set to 161, then investigate whether longer generation requires full training or can use DiffSynth LoRA training. Done should establish and document the supported approach for longer videos.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Feature
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.