Rhythm Feature Extractor ⚡🅡🅞🅣🅘

AudioFeatureExtractor has an onset_strength method for catching individual hits, but if what you actually want is beat-synced - an effect that lands on the downbeat, or a pulse that tracks tempo - this is the more purpose-built node for it. It analyzes the rhythmic structure of a track rather than just its raw loudness or brightness, and turns that into a FEATURE you can wire into any of the pack's reactive targets.

The eight extraction methods

extraction_method covers a genuinely wide range of rhythmic information, and the author's own tooltip descriptions make clear when to reach for each: beat_locations marks exact beat timings - the choice for precise beat-synced effects. tempo gives overall speed, for effects that should scale with how fast or slow the track is rather than react per-beat. onset_strength captures how strong each beat hits, good for impact effects. beat_emphasis highlights stronger beats specifically, useful for accenting the main hits without reacting to every minor one. syncopation detects off-beat rhythms - a more advanced, textural choice for complex or unusual rhythmic material. rhythm_regularity measures how steady the beat is, a stability signal rather than an impact one. down_beats marks the first beat of each measure (strong emphasis), while up_beats marks the others (lighter emphasis) - a genuine distinction if your effect should feel different on beat one versus the rest.

time_signature (default 4, for common 4/4 time) is what makes the down-beat/up-beat distinction actually correct - it tells the node how many beats make up a measure, so it knows which beat is "the one."

Like the pack's other audio extractors, frame_count at 0 auto-calculates from the audio's length and frame_rate; set it explicitly and the extracted feature gets interpolated to that target length instead.

The inputs and outputs that matter

• audio (required, AUDIO) - the source.
• extraction_method - one of the eight rhythmic measures above.
• time_signature (default 4) - beats per measure; critical for down_beats/up_beats to be meaningful.
• frame_rate, frame_count (default 0, auto) - how the rhythm timeline maps onto video frames.
• width/height - sizes the internal feature tensor; match your pipeline's resolution.

Output is a single FEATURE - wire it into any Flex target's reactive input, or FeatureToMask if you just want the rhythm as a plain mask.

How to install it

ComfyUI Manager: search "RyanOnTheInside", install, restart. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ryanontheinside/ComfyUI_RyanOnTheInside
cd ComfyUI_RyanOnTheInside
pip install -r requirements.txt

then restart. Audio analysis dependencies come in via requirements.txt with the rest of the pack - no separate model download for this node.

Common issues & troubleshooting

Downbeats land on the wrong beat. This is almost always time_signature mismatched to your actual track - a waltz in 3/4 analyzed with time_signature left at the default 4 will misidentify which beat is the downbeat throughout. Set it to match the actual meter of your source.

Feature looks flat on a track with an obvious beat. Try a different extraction_method before assuming the analysis failed - syncopation and rhythm_regularity in particular are subtler, texture-level measures that can read as nearly flat on very straightforward, four-on-the-floor material, while onset_strength or beat_locations on the same audio would show clear activity.

Tempo detection seems off on a track with tempo changes. tempo extraction assumes a reasonably steady pulse to lock onto - a track with significant tempo drift or rubato (a live, unquantized performance, for instance) can confuse it. beat_locations tends to hold up better than tempo on that kind of material since it tracks individual hits rather than inferring one global speed.