H3 Audio Trim Start

One frame. That's the whole job. H3AudioTrimStart chops a number of seconds off the front of an audio clip, and its default value is 0.04167 - which is exactly 1/24th of a second, exactly one frame at 24fps. If that number looks weirdly specific, it's because it's load-bearing.

Here's the situation it exists for. When H3 Multishot chains two shots together, shot 2 opens with a copy of shot 1's closing frame - that's the relay that makes the join seamless. But that duplicated frame has to come out of the final video, and if you trim the picture without trimming the audio the same amount, your lip sync drifts by a frame at every seam. Speech is half the model in H3, and a chain full of one-frame audio offsets is how you get dialogue that looks almost right but never quite lands. This node trims the audio by the same amount, in lockstep, so the pair stays exact.

How it works

Two inputs, one output. audio is the clip, seconds is how much to cut off the front (0 to 10 seconds, default 0.04167). Out comes the trimmed AUDIO. There's nothing clever in the implementation - it's a front trim, pure and simple - and that's the virtue. It's the utility node you wire up when you're doing your own seam math instead of letting the bundled samplers handle it.

When you'd actually use it

Mostly you won't, because the samplers in this pack do their own seam trimming internally. But if you're building a graph from stock H3 nodes, or you're concatenating shots with ComfyUI's own Concat nodes, or you trimmed a video's duplicate boundary frame with a video tool and your audio is now half a frame long - this is the fix. It's also a fine general-purpose "cut the silence at the head of a clip" tool, since 0 to 10 seconds covers most dead-air cleanup.

A practical note: keep fps at 24 in this pack's chains. The README is explicit that other frame rates audibly shift voice accents, and if you do run something off-spec, remember your seconds value and your video trim have to agree on the frame rate. Trim 2 frames at 30fps (0.0667s) from the video and 2 frames at 24fps (0.0833s) from the audio and you've just invented a lip-sync error the node was designed to prevent.

Install

Part of the H3 Multishot pack:

cd ComfyUI/custom_nodes
git clone https://github.com/jlucasmcrell/ComfyUI-H3-Multishot

or ComfyUI-Manager → search H3 Multishot. Restart ComfyUI; needs v0.30.0+. No extra dependencies - the pack deliberately declares none, since everything it imports ships with ComfyUI.