Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

Is it possible to fine-tune Minimax H3 with silent videos?

Open
  • Is something wrong?
#1,576 1 comment 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
4/5
Estimated time
3-5 days
Newbie friendliness
48/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Quiet
Tech stack
python, pytorch
Domain
audio-video-rtc, machine-learning

Research direction

Start with examples/minimax_h3/model_training/train.py and trace the data inputs into diffsynth/diffusion/loss.py, especially FlowMatchSFTMiniMaxH3AudioVideoLoss. Reproduce the command without input_audio and inspect where audio_input_latents is expected. Done means silent-video fine-tuning no longer raises the reported KeyError and the training path has a defined outcome for missing audio.

Written by the indexing model from the issue text.

Description

Here is my cmd:

accelerate launch examples/minimax_h3/model_training/train.py \
  --dataset_base_path data/diffsynth_example_dataset/minimax_h3/MiniMax-H3-FL2VA \
  --dataset_metadata_path data/diffsynth_example_dataset/minimax_h3/MiniMax-H3-FL2VA/metadata.csv \
  --data_file_keys "video" \
  --extra_inputs "input_image" \
  --height 480 \
  --width 832 \
  --num_frames 124 \
  --dataset_repeat 1 \
  --model_id_with_origin_paths "MiniMax/MiniMax-H3:FL2VA/text_encoder/model*.safetensors,MiniMax/MiniMax-H3:FL2VA/video_vae/source/model.safetensors,MiniMax/MiniMax-H3:FL2VA/audio_vae/model.safetensors" \
  --learning_rate 1e-4 \
  --num_epochs 1 \
  --remove_prefix_in_ckpt "pipe.dit." \
  --output_path "./models/train/MiniMax-H3-FL2VA-split-cache" \
  --lora_base_model "dit" \
  --lora_target_modules "qkv_proj,out_proj" \
  --lora_rank 32 \
  --use_gradient_checkpointing \
  --task "sft:data_process"

# FL2VA - stage 2 (train)
accelerate launch examples/minimax_h3/model_training/train.py \
  --dataset_base_path ./models/train/MiniMax-H3-FL2VA-split-cache \
  --data_file_keys "video" \
  --extra_inputs "input_image" \
  --height 480 \
  --width 832 \
  --num_frames 124 \
  --dataset_repeat 100 \
  --model_id_with_origin_paths "MiniMax/MiniMax-H3:FL2VA/transformer/model*.safetensors" \
  --learning_rate 1e-4 \
  --num_epochs 5 \
  --remove_prefix_in_ckpt "pipe.dit." \
  --output_path "./models/train/MiniMax-H3-FL2VA-split" \
  --lora_base_model "dit" \
  --lora_target_modules "qkv_proj,out_proj" \
  --lora_rank 32 \
  --use_gradient_checkpointing \
  --find_unused_parameters \
  --task "sft:train"

I remove input_audio and get this error:

File "/home/test/project/DiffSynth-Studio/diffsynth/diffusion/loss.py", line 78, in FlowMatchSFTMiniMaxH3AudioVideoLoss
    audio_noise = torch.randn_like(inputs["audio_input_latents"])
                                   ~~~~~~^^^^^^^^^^^^^^^^^^^^^^^
KeyError: 'audio_input_latents'

How to fine-tune Minimax H3 using Videos without audio?

Dominant language
Python
Stars
13.2k
Forks
1.3k
Avg merge
22h 15m
Merged PRs (30d)
31

Getting set up

This project ships no dev container, Dockerfile or contributing guide, so setting up is up to you: start from its README, and see our first-contribution guide for the general steps.

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.

More from modelscope/DiffSynth-Studio

All issues in modelscope/DiffSynth-Studio

Similar issues

More Python issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.