🎵→🎬 Wan Audio-to-Video

Wan Audio-to-Video is not a model and not Wan. Open the source and you'll find no diffusion, no weights, no AI at all: it's a procedural audio visualizer that draws shapes based on your audio's samples and FFT. The "🎵→🎬" in the name is the honest part. The "Wan" is branding left over from the pack's fantasy theme.

Here's the thing though - it works, it's instant, and there's no model to download. You feed it a ComfyUI AUDIO input (say, from the pack's Studio42 Audio Loader, or any audio node that outputs an AUDIO dict with a waveform) and it renders up to 60 frames of visualization at 30fps - roughly two seconds per render pass.

The styles

• Waveform - a polyline of the raw audio samples around the vertical center.
• Spectrum - FFT magnitude bars, the classic spectrum analyzer look. Usually the most legible.
• Bars - 32 amplitude bars, simpler and chunkier than the FFT version.
• Particles / Abstract - a radial burst where each point's distance from center follows the audio amplitude. Pretty, less readable.

Colors come from color_scheme - LCARS Orange, Neon Blue, Rainbow, Monochrome - which at least commits to the pack's theme. width (256–1920) and height (256–1080) set the canvas.

What you set

• audio - the AUDIO input. It's read as a waveform tensor plus sample_rate.
• style - one of the five above.
• width/height, color_scheme - as above.

Output is a batch of IMAGE frames (output), which you feed to a video save/output node like any generated clip.

Where it fits

This is a fun, cheap way to make music-reactive placeholder footage without paying for a commercial visualizer or a heavy generative model. Pair it with the pack's Studio42 Audio Loader and Mixer and you've got a full audio→visual pipeline with zero model weight. The honest limits: output is capped at ~60 frames (two seconds at 30fps) per render, it's per-render rather than continuous, and the visuals are geometric - don't expect it to "understand" the music.

Installing it

Same pack, 24oiduts-ComfyUI:

• ComfyUI Manager: search 24oiduts, install, restart.
• Manual:

  cd ComfyUI/custom_nodes/
  git clone https://github.com/GeekyGhost/24oiduts-ComfyUI
  pip install -r 24oiduts-ComfyUI/requirements.txt

  then restart.

It needs opencv-python for the drawing helpers (and falls back gracefully if absent - output just gets sparser). The GitHub README is a stale Studio42 template; the actual docs are LCARS_README.md in the repo.

Troubleshooting

• No bars/waveform drawn - OpenCV missing, so the draw helpers skip. pip install opencv-python.
• Very short output - the node caps at 60 frames / ~2s per run. Feed it a section of audio, not an hour-long track.
• Audio input refused - make sure the upstream node really outputs the ComfyUI AUDIO dict (waveform + sample_rate), which the loader in this pack does.

It's a visualizer with a misleading name, but honestly labeled about what it does. For quick reactive-placeholder footage with zero model downloads, it's one of the most fun nodes in the pack.