MiniMax H3 Silent Audio Slice (Internal)

The silent sibling of Locked Audio Slice, and the pair is easier to understand together: one extracts the interval of your recording for a segment, the other manufactures the interval of nothing.

Why would you want nothing? Because in this pack, "video audio is disabled" isn't handled by ignoring audio. The changelog spells out the policy: disabling video original audio "fixes the AV audio stream and final master to silence instead of asking H3 to generate replacement sound." That's a real choice with real consequences - H3's defining feature is jointly generated picture and sound, so if you mute the source and let it run normally, it will happily invent a soundtrack for your reference video. If you're doing a character replacement or an edit where the audio must come from elsewhere, invented sound is exactly what you don't want. So the pack feeds it silence, on purpose, at zero denoise.

How it works

There's no cleverness here, and that's a feature. It takes one segment's material plan, resolves the segment's frame length from the plan's length or an output_frames override, and returns a zero waveform: one channel, 44100 Hz, round(frame_count / 24 * 44100) samples with a floor of one. No padding logic, no asset lookup, no trim offset - because there's no source to offset into. Where Locked Audio Slice has a branch for standalone-versus-timeline assets, this node has one code path.

It fails the same way its sibling does when the plan can't yield a duration: "The silent-audio interval has no target frames" means the plan isn't a real per-segment plan.

In the shipped graphs the output goes into VAEEncodeAudio with the H3 audio VAE and then into Lock Audio Latent, which pins the audio stream at zero denoise. So the segment samples normally on the video side and holds a silent audio stream throughout. That's how you get a segment whose picture is generated but whose sound is guaranteed empty.

Inputs and outputs

Required: plan - one segment's MINIMAX_H3_TIMELINE_PLAN. Per-segment, not the whole timeline; that's what makes this a slice rather than a master.

Output: Silent Audio (AUDIO), an empty waveform of exactly the segment's duration. One sample minimum, so downstream latent encoding always has something legal to chew on even for a pathological length.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/Songssx/ComfyUI-MiniMaxH3-TimelineDirector.git

Restart, search MiniMax H3. Nothing to pip install - the pack declares an empty dependency list and runs on what ComfyUI already brings. Like the rest of the (Internal) family, it's dev-only and stays out of the add-node menu unless dev mode is enabled in ComfyUI's settings. Pack prerequisites: a recent ComfyUI with the native H3 nodes, the H3 Ref2VA model, CLIP, video VAE and audio VAE, and Python 3.10+.

Common issues

"The silent-audio interval has no target frames." You passed a plan with no segment length - most often the whole-timeline Material Plan instead of one segment's plan, or a plan whose generation window was never set.

The segment is silent but neighbouring segments aren't. Working as intended. This is per-segment policy from the plan's assignments; each iteration resolves its own. If you wanted the whole video silent, that's the master node's job.

You hear generated sound despite muting. Something upstream isn't using this path - check that the plan actually reports muted video audio, and remember that an explicitly uploaded locked audio asset takes priority over the muted path. Both are configured on the timeline, not on this node.

Segment came back shorter than expected. The duration here follows the segment's frame length, which comes from the plan. If the window snapped a frame or two in the planner, the silence follows it exactly - which is the behaviour you want, since a longer silent track than the picture is its own kind of bug.