modelscope / modelscope/DiffSynth-Studio

Is it reasonable to train the low_noise part and high_noise part of Wan2.2 separately?

Open
#734 1 comment 0 reactions 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

I think training the two parts of parameters separately will lead to incompatibility during inference.
I guess the original training of Wan2.2 should have been conducted with both parts together?

examples/wanvideo/model_training/lora/Wan2.2-I2V-A14B.sh

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

The referenced entry point is examples/wanvideo/model_training/lora/Wan2.2-I2V-A14B.sh; read it first to see how the low_noise and high_noise parts are selected during training. Check the Wan2.2 training assumptions and determine whether separate training is supported or joint training is required; done means a confirmed compatibility answer and any needed documentation update.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, shell
Domain
ai, 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.