MultiTrack Audio Output

The MultiTrack editor in ComfyUI-Easy-Media is all about assembling audio, video, and task segments in one timeline. MultiTrack Audio Output is the node you reach for when you need to pull the sound out of that timeline and hand it to a model - most often for speech-driven generation, where the audio is the whole point rather than a side channel.

Its description sums up the intent: merge all audio tracks and output the first two tracks, in full or cropped, for S2V (speech-to-video). "S2V" here is the dubbing/avatar world - MiniMax H3's speech modes, Bernini S2V, talking-head pipelines where a clip of dialogue should drive the lips and the motion.

What it gives you

Feed it the TRACKS_INFO from a MultiTrack editor and you get five outputs:

• combine_audio - every audio track merged into one.
• audio_0 and audio_1 - the first and second audio tracks individually.
• audio_0_start and audio_1_start - integer start offsets (in frames or samples, matched to the timeline's frame rate) for each of those first two tracks. Handy when you crop and need to know where the clip starts relative to the timeline.

Two inputs control the behavior. mode flips between default and crop:

• default - you get the full tracks. Simplest, use this first.
• crop - the node crops the first two tracks to the range of the task segment you pick, so you can feed a per-task slice of dialogue to an S2V model instead of the entire audio track.

task_index selects which task segment's range the crop mode uses (zero-based, matching the task segments you see in the MultiTrack editor). There's also an optional audio input - if you're running a slot-backed timeline where the editor materializes media immediately, that input is how the actual audio gets in; when the timeline defers loading, the node pulls the audio from the track data itself.

Why it exists

Without this node you'd be dragging audio out of a list, trimming it by hand, and wiring merges - the pack's whole pitch is that the multitrack replaces that choreography. In a dubbing workflow the typical flow is: MultiTrack editor → this node → MiniMax H3 r2va or a Bernini S2V conditioning node → generate → MultiTrack Add Subtitle To Video on the way out. The audio_0/audio_1 split exists because a two-speaker conversation needs each voice addressed separately, which is exactly the sequential two-speaker setup some S2V models expect.

Install and gotchas

It ships in ComfyUI-Easy-Media, installable via ComfyUI Manager (search "Easy-Media") or:

cd ComfyUI/custom_nodes
git clone https://github.com/yolain/ComfyUI-Easy-Media.git

Restart ComfyUI after cloning. FFmpeg on your system is the one hard prerequisite the README calls out - audio handling leans on it heavily.

The trap most people hit: forgetting which mode they're in. In crop mode with a task_index that points past the last task segment, the range is empty and you get silence. Also remember the outputs are the raw timeline audio - if your audio is misaligned with the video track, fix it in the MultiTrack editor, because this node won't (and shouldn't) resync anything. And don't expect it to do anything useful without a TRACKS_INFO upstream: it's a pure consumer of the multitrack data, not a standalone audio loader.