Audio visualizer nodes for ComfyUI, replicating audio-visualizer.com: load audio, pick a style, tune the look, get an image batch (+ alpha mask) you can turn into a video — including transparent overlays.
Built on the ComfyUI V3 node API (io.ComfyNode + io.Schema, registered
via ComfyExtension/comfy_entrypoint). Requires a ComfyUI version with
comfy_api.latest (0.33 tested).
Audio Visualizer Analyze — AUDIO → AUDIO_FEATURES
Precomputes one spectrum + waveform slice per video frame, mimicking the browser's
Web Audio AnalyserNode (Blackman window, dB scaling −100..−30, temporal smoothing).
| Input | Matches site control |
|---|---|
fps |
video frame rate (site renders live at ~60) |
fft_size |
"Detail (FFT)" — 32..2048 |
smoothing |
"Smoothing" — 0.8 default |
sensitivity |
"Sensitivity" — 1.4× default |
Audio Visualizer Render — AUDIO_FEATURES → IMAGE + MASK + fps
Renders the frames. style covers all 8 site styles: bar, wave, line,
radial_bars, mirror_bars, mountain, blob, siri. Also: width/height,
color (hex), background (transparent/dark/light/custom), bar_width, gap,
line_width, amplitude.
Analysis and rendering are separate nodes on purpose: the FFT pass is cached by ComfyUI, so switching styles or colors re-runs only the (cheap) render node.
The Render node embeds a live canvas preview (js/audio_viz_preview.js) — press ▶ Play on the node and the browser analyzes the connected audio with a Web Audio AnalyserNode (the same model the offline analysis mimics) and draws the selected style in realtime. Tweaking style/color/bar width/etc. updates the preview instantly, without running the graph. The audio file is resolved by walking upstream Render ← Analyze ← LoadAudio.
The preview is a draft (analyser smoothing runs at display rate, not at the export fps), so the exported video can differ subtly — same look, not frame-identical.
LoadAudio ─→ Audio Visualizer Analyze ─→ Audio Visualizer Render ─→ CreateVideo ─→ SaveVideo
└────────────────────────────────(audio)──────────↗
- Solid background video: set
backgroundto dark/light/custom, wireimagesinto CreateVideo along with the original audio. - Transparent overlay (site's WebM-with-alpha): keep
background= transparent and use theMASKoutput — e.g. composite over footage with ImageCompositeMasked, or JoinImageWithAlpha → SaveWEBM for an alpha WebM.
/Users/rob/Documents/ComfyUI/.venv/bin/python tests/smoke.py
Writes one PNG per style to tests/out/ and exercises both nodes end-to-end.
The repo is symlinked into the local install's custom_nodes
(/Users/rob/Documents/ComfyUI/custom_nodes/ — the server launches with
--base-directory /Users/rob/Documents/ComfyUI, so that's the only dir it
scans; server on port 8000) — restart ComfyUI to pick up code changes.
No extra dependencies: numpy, torch, and Pillow all ship with ComfyUI.