AUDIO Emotion Visualizer @ vrch.ai

The AUDIO Emotion Visualizer @ vrch.ai is what makes the pack's emotion detection look like something. The Music to Emotion Detector scores a track and hands you RAW_DATA; this node takes that JSON and renders it as three actual images - a radar chart of the mood scores, a word cloud of the top moods, and a valence-arousal scatter plot. Wire the images into a Web Viewer and you've got a live "mood board" that redraws as the music changes.

That's the whole appeal in one sentence: it converts abstract model outputs into something you can look at, stream to a browser, or even feed back into a video pipeline. It's the display half of the pack's music emotion example workflow, and it doubles as a debugging tool - a radar chart tells you in one glance whether the detector is actually tracking the song or just staring at a flat line.

How it works

You feed raw_data - the JSON from the Music2Emotion Detector - in one side, and three IMAGE outputs come out the other:

• MOODS_RADAR_IMAGE - mood scores plotted as a radar chart. radar_top_k (3–12) picks how many moods to show, radar_normalize scales them, and the toggles control whether labels and values are drawn.
• MOODS_WORDCLOUD_IMAGE - the moods as a word cloud sized by score. wordcloud_top_k, wordcloud_size_scale, and a handful of layout options (center_spiral, concentric_rings, sequential_flow, …) control the look.
• VALENCE_AROUSAL_IMAGE - the track's position in valence-arousal space. The va_ options tune the gridlines, axis labels, and the point color.

Then there's a long list of styling inputs: image_width/image_height, background_color, font_color, font_size, and a shared theme system (pastel, ocean, sunset, forest, neon, mono-blue, mono-gray) that applies to each chart independently - radar_theme, wordcloud_theme, va_theme. If the defaults feel like a default, this is one of those nodes where ten minutes of theme-swapping is genuinely fun. debug logs the render.

Installing it

Standard pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/VrchStudio/comfyui-web-viewer
pip install -r comfyui-web-viewer/requirements.txt

Or install "ComfyUI Web Viewer" via ComfyUI Manager and restart. No models to fetch - but remember the upstream detector needs the optional Music2Emotion submodule, and without it there's no raw_data to visualize.

Practical notes

• This node expects the Detector's JSON, not arbitrary audio data. Plugging a Mic Loader's RAW_DATA into it won't render anything meaningful - the field names won't match. Keep the chain Detector → Visualizer.
• The images are plain IMAGE tensors, so they compose like any other image in ComfyUI - save them, upscale them, put them in a batch, or stream them through a Web Viewer for the realtime effect.
• For a snappier live update, keep radar_top_k and wordcloud_top_k modest. At the max settings the charts take noticeably longer to redraw every generation cycle, which matters when your workflow is re-running on instant queue.