Audio Visualizer
Turn a still image and audio into a spectrum visualizer video with bars or LED dots, using FFmpeg's libraries through PyAV.
Audio Visualizer for ComfyUI
Turn a still image and audio into an MP4 or WebM with a bar or dots spectrum, or just the still image. Insert your own image and use the mock preview to adjust the style.

Install
- ComfyUI-Manager: search for Audio Visualizer (Registry ID
comfyui-audiovisualizer) and install it. - comfy-cli:
comfy node install comfyui-audiovisualizer. - Manually:
git clone https://github.com/bno84/ComfyUI-AudioVisualizerintoComfyUI/custom_nodes/.
Then restart ComfyUI and refresh the browser. Requires ComfyUI 0.38.0 or newer. It uses PyAV (av>=17), already required by that ComfyUI version, and ComfyUI's NumPy. No GPU or separate FFmpeg executable is needed.
Manager installations install the pack's Python requirements automatically.
For a manual installation, install requirements.txt using the same Python
environment that runs ComfyUI. Standard PyAV wheels include FFmpeg libraries
for Windows, macOS and Linux; system or custom builds may have different
features. The node checks only the selected style and codecs, and names
missing filters/encoders in its error. Installing an ffmpeg executable
does not add features to PyAV's libraries. With conda or another managed
environment, use its package manager to manage PyAV/FFmpeg rather than
overlaying a pip installation; codec availability depends on that build.
Use
Add 🎵 Audio visualizer video (still + spectrum) from audio/visualize
and connect an image and audio. It saves to ComfyUI's output folder and
returns a VIDEO, with the workflow embedded in the file.
Optionally connect text to title, artist, comment and lyrics. Each one connected and not empty is saved in the video's metadata. Some players, such as Dragon Player, show the title; VLC does not.
- Visualization: (none) (still image + audio only), bars, or dots, color, band height, gain, glow, and trails. Spectrum controls hide in (none) mode; dot spacing appears in dots mode.
- Band height: the spectrum band's share of the video height, 0.1–0.9 (default 0.35).
- Gain: the height of the loudest moment. The audio is measured before rendering, so 1.0 makes the loudest bar or dot column just reach the top of the band, whatever the volume. The default 1.3 lets the loudest peaks clip; the range is 0.25–2.
- Vignette: edge darkening from 0 (off) to 1 (strongest). The default is 0 (off); 0.4 reproduces the original look. Workflows without this control also default to off; explicitly saved strengths are preserved. The preview uses the same edge-darkening falloff as the render.
- Preview: the node shows the connected still with the band at its height and a mock spectrum peaking at the chosen gain, to check placement before rendering. The spectrum shape is made up; only the real render uses the audio.
- Fades: visual fade-in and fade-out from/to black, each 0–3 seconds. Zero disables the fade. Defaults are 1.5 seconds in and 2.5 seconds out, including for older workflows. Fades are capped at the rendered duration for short clips and can overlap. Audio is unchanged.
- Timing: 30/60 FPS, start, and duration in seconds. Duration 0 uses the rest of the audio.
- Codecs: H.265 (CRF 18), H.264 (CRF 22), or VP9 (CRF 20, libvpx-vp9); FLAC (default), AAC (192 kbps), or Opus (128 kbps). Each codec keeps its own settings.
- Container: MP4 (default, FastStart) or WebM. Choose VP9 + Opus for WebM; other combinations are rejected before rendering. WebM writes its seek index at the front.
The filename includes the style automatically, e.g.
video/song_visualizer_bars_00001_.mp4. Disable enabled to skip rendering.
Bars and dots measure the audio through library filters before rendering.
Quiet audio can need multiple measurement passes; (none) skips analysis.
Rendering feeds bounded audio chunks and frames directly to the encoders,
without temporary raw image/audio files or a full-song image batch. The video
is published only after encoding closes successfully; cancellation removes
the temporary output. A hard process kill can leave a .audio-viz-*
file in the output folder; it can be deleted after that process has stopped.
Existing workflows retain their MP4 output and codec settings. The FFmpeg libraries in a different PyAV version can produce slightly different pixels. Opus in MP4 can expose encoder padding when decoded to raw samples, although the recorded audio duration is preserved.
Workflows saved with 0.1 keep their settings, but gain resets to 1.3: it used
to be a raw volume. API prompts from 0.1 need band_height and a gain between
0.25 and 2.0.
Tests
python3 -m pip install -r requirements.txt numpy
python3 -m unittest discover -s tests
node --test tests/*.mjs
Render tests use real PyAV filters and encoders and block subprocess rendering. CI runs them on Linux, Windows and macOS, including the minimum PyAV version.