FL Audio Segment Extractor

Given a full track's beat positions and a starting beat plus a beat count, this node slices out exactly that span of audio - and, critically, computes the exact frame count that span corresponds to at your target framerate. That pairing matters more than it sounds: if you cut audio by raw seconds and cut video by separate frame-count math, rounding error compounds over a long edit and the two creep out of sync. Deriving both from the same beat positions keeps them locked to a single source of truth.

How it works

You give it beat_positions (from FL_Audio_BPM_Analyzer upstream), a start_beat to begin at, and a beat_count for how many beats long the segment should be. It returns the corresponding audio slice plus the exact frame_count and end_frame that span occupies at your chosen fps - no separate manual conversion between "beats" and "frames" required anywhere downstream.

The inputs and outputs that matter

Required: audio, beat_positions, start_beat (0–10000, default 0 - 0-indexed), beat_count (1–32, default 4 - four beats is one bar in common 4/4 time, which is why that's the default).

Optional: fps (1–120, default 30); start_frame (0–1,000,000, default 0 - an offset useful when you're stitching multiple segments together and want this segment's frame numbers to continue where a previous one left off, instead of restarting at 0).

Outputs: audio_segment (AUDIO), frame_count (INT), end_frame (INT).

How to install it

ComfyUI Manager: search ComfyUI_Fill-Nodes, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI_Fill-Nodes

then restart. The audio-analysis dependency (librosa, for the BPM analyzer this node relies on upstream) lives elsewhere in the pack - this specific node is comparatively lightweight, doing slicing and arithmetic rather than analysis.

Common issues & troubleshooting

Segment lands on the wrong part of the song. beat_positions has to come from an actual BPM analysis of this exact audio file - if you're reusing beat data from a different track, or from a re-edited version of the same track, the beat positions won't line up and the slice will be wrong even though nothing errors.

Stitching multiple segments for a full edit. Use start_frame on each subsequent extractor call to keep frame numbering continuous across segments - otherwise every segment restarts its frame numbering at 0, which breaks any downstream logic assuming a single continuous timeline. If you've already built a full shot sequence rather than calling this node manually per segment, FL_Audio_Shot_Iterator is the node built to walk that sequence instead.

beat_count capped at 32. That's a hard ceiling in the schema - if you need a longer continuous segment than 32 beats, you'll need to chain multiple extractor calls (using start_frame to keep them contiguous) rather than trying to push past the limit in one call.