Nodes/ComfyUI-ALLinONE-MinimaxH3/H3 Audio Trim (internal)
ComfyUI Node

H3 Audio Trim (internal)

Keep H3 reference audio inside its 2–15 second window

By LeonQ8·Created 10 days ago·Updated about 17 hours ago· 273
H3 Audio Trim (internal)
  • audio
  • audio
trim_seconds5.0

MiniMax H3 is one of the few open video models that generates audio natively - the soundtrack is created jointly with the picture, not bolted on after. But that audio plumbing comes with a strict input contract: reference audio clips are specced at 2–15 seconds, and an audio reference longer than the target video is out-of-distribution for the model's packed layout. Feed it a 40-second clip and you're asking it to work with something it wasn't trained to see. H3 Audio Trim (internal) is the guardrail that keeps ref audio inside that window.

It's about as simple as a node gets. Two inputs:

  • audio (AUDIO) - the reference track.
  • trim_seconds (FLOAT, default 5, range 0.5–15) - the target duration.

One output: the trimmed audio.

Mechanically it just cuts the waveform to trim_seconds × sample_rate samples. The clamp is what makes it safe: whatever you type, it's capped at 15 seconds and floored at half a second, so you can't accidentally tell it to produce something H3 can't consume. If the incoming audio is already shorter than the target, it passes through untouched - no upsampling, no stretching, no surprises. That "shorter passes through" behavior is worth knowing, because it means the node is a cap, not a resampler; it will never invent audio you didn't give it.

You won't add this node to a canvas yourself - it's generated inside the pack's R2V, Audio Drive, and Mask workflows whenever the frontend needs reference audio normalized, which is why it carries the "(internal)" suffix. But the lesson travels: H3's reference spec isn't arbitrary, and if you're hand-building H3 graphs outside this pack, the same 2–15 second rule applies to any audio you wire into the conditioning yourself. Audio that "doesn't work" in H3 is often just audio that's too long.

Install-wise there's nothing to do - it ships with the ALLinONE-MinimaxH3 pack alongside the rest of the internal nodes, and it needs no models of its own beyond the H3 audio VAE (minimax_h3_audio_vae_fp32.safetensors) that the main pipeline already loads. It's the kind of node you'll never notice until it saves you from a baffling failure, and then you'll be glad it's there.

CategoryOne Node

Inputs (2)

NameTypeDefaultDescription
audioAUDIO
trim_secondsFLOAT5.00.5–15

Outputs (1)

NameTypeDescription
audioAUDIO