n0p/mlx-video
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add librosa dependency and enhance A2V documentation with additional pipeline options

Browse Source
This commit is contained in:
Prince Canuma
2026-03-16 02:02:13 +01:00
parent 6f6105b715
commit decb3eb9e5
2 changed files with 14 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
README.md
View File

@@ -88,15 +88,23 @@ uv run mlx_video.generate --pipeline dev --prompt "Waves crashing" --image beach
### Audio-to-Video (A2V)
Generate video conditioned on an input audio file. The audio is encoded to latent space and frozen during denoising — the transformer's cross-attention reads the audio signal to guide video generation.
Generate video conditioned on an input audio file. Works with all four pipelines. The audio is encoded to latent space and frozen during denoising — the transformer's cross-attention reads the audio signal to guide video generation.
```bash
# A2V - generate video from audio
# A2V - distilled (default, fastest)
uv run mlx_video.generate --audio-file music.wav --prompt "A band playing music"
# A2V with dev pipeline
# A2V - dev (single-stage with CFG)
uv run mlx_video.generate --pipeline dev --audio-file ocean.wav --prompt "Ocean waves"
# A2V - dev-two-stage (dev + LoRA refinement)
uv run mlx_video.generate --pipeline dev-two-stage --audio-file music.wav \
--prompt "A band playing music" --model-repo prince-canuma/LTX-2-dev
# A2V - dev-two-stage-hq (highest quality)
uv run mlx_video.generate --pipeline dev-two-stage-hq --audio-file music.wav \
--prompt "A band playing music" --model-repo prince-canuma/LTX-2-dev
# A2V + I2V (audio + image conditioning)
uv run mlx_video.generate --audio-file rain.wav --image forest.jpg --prompt "Rain in forest"
@@ -104,6 +112,8 @@ uv run mlx_video.generate --audio-file rain.wav --image forest.jpg --prompt "Rai
uv run mlx_video.generate --audio-file song.mp3 --audio-start-time 30.0 --prompt "Concert"
```
> **Note:** `--audio-file` (A2V) and `--audio` (generate audio) are mutually exclusive. Supported formats: WAV, FLAC, MP3, OGG, and video files with audio tracks.
### Audio-Video Generation (experimental)
Generate synchronized audio alongside video from scratch:

1
pyproject.toml
View File

@@ -21,6 +21,7 @@ dependencies = [
"Pillow>=10.3.0",
"mlx-vlm",
"rich>=14.2.0",
"librosa>=0.10.0",
]
license = {text="MIT"}
authors = [