File size: 551 Bytes
6febba8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
# Vendored lot/ + third_party/musubi_tuner subset pulls these on the load path:
#  - av            (minimax_h3/media.py + dataset/*, module-level)
#  - opencv        (dataset/media_utils.py via cv2_compat — headless, no GUI deps)
#  - accelerate    (init_empty_weights in model.py / video_vae.py)
#  - imageio(+ffmpeg) for two_track_h3.save_clip's mp4 writer
# gradio, spaces, huggingface_hub and torch are preinstalled on ZeroGPU — not pinned here.
accelerate
safetensors>=0.8.0
numpy
pillow
tqdm
av
opencv-python-headless
imageio
imageio-ffmpeg