Nodes/ComfyUI-MiniMax-Creator/MiniMax H3 Clip Audio
ComfyUI Node

MiniMax H3 Clip Audio

The clip's opening or closing seconds — the audio side of a seam around supplied footage

By roadmaus·Created 17 days ago·Updated about 18 hours ago· 58
MiniMax H3 Clip Audio
    • AUDIO
    clip_data
    seconds1.0
    attail

    Say a real clip you shot or downloaded is sitting in the middle of your H3 timeline, with generated shots before and after it. The shot running into the clip and the shot running out of it both need audio context, or the sound will audibly reset at the cut. MiniMax H3 Clip Audio is the node that supplies it: the first or last seconds of a supplied clip's soundtrack, for the seam beside it.

    It's the audio sibling of MiniMaxH3ClipFrames, and it follows the same laziness principle. Sound can't be handed over as a single frame - a single sample says nothing about a room - so it needs a stretch of seconds instead. And since the clip was never generated (so it has no spill to read back), the node pulls what it needs from the clip's file on demand, and only when a neighbouring generation actually wants it.

    The role it plays is the same one MiniMaxH3PassAudio plays between generated passes, and the naming tells you which case you're in: Pass nodes read from a decoded pass's spill, Clip nodes read from a supplied file. This one is how a supplied clip's room tone, dialogue tail, or music carries across a cut into or out of the generated shots around it - the audio half of what makes a chained piece feel like one continuous world instead of a string of clips.

    The inputs that matter

    • clip_data - the JSON blob describing the supplied clip, same format as the rail's clip cards.
    • seconds - default 1, range 0.1–4. How much of the soundtrack the seam inherits. Short on purpose: enough to carry a room, not enough to drag the whole previous shot's soundscape along.
    • at - tail or head, default tail. The tail is what a generation after the clip continues from; the head is what a generation before it ends on.

    Output is an AUDIO tensor feeding the segment's prev_audio / next_audio inputs in the graph the Creator builds for a chained timeline.

    Installing

    It's internal, dev-only, and ships with the pack:

    cd ComfyUI/custom_nodes
    git clone https://github.com/roadmaus/ComfyUI-MiniMax-Creator
    

    Restart ComfyUI, no pip install, standard H3 weights in the usual folders. If you've ever built a piece with real footage in the middle and heard the sound kick back to silence at the edit point, this is the node that quietly prevents it - one of the small, easily overlooked parts of the pack that makes multi-shot H3 feel like editing rather than stitching.

    CategoryMiniMax/internal

    Inputs (3)

    NameTypeDefaultDescription
    clip_dataSTRING
    secondsFLOAT1.00.1–4
    atCOMBOtail2 options: head, tail

    Outputs (1)

    NameTypeDescription
    AUDIOAUDIO