VELVET VICE MiniMax H3 — Native Audio Gate

There are two different "mute" switches in the Velvet Vice MiniMax H3 pack, and confusing them is how you end up with a video that plays in the ComfyUI preview but comes out of the encoder silent. VelvetViceMiniMaxH3AudioDecodeGate decides whether audio gets decoded; VelvetViceMiniMaxH3AudioGate is the one right before encoding that decides whether decoded audio actually makes it into the final file.

This node is the simpler of the two, which is why it's easy to overlook. It takes already-decoded AUDIO on one side and a boolean on the other. If the boolean is true, the audio passes through untouched and the node reports WITH SOUND · Native H3 audio attached to final video. If it's false, you get no audio and the status string flips to MUTED / VIDEO ONLY · final video has no audio track.

Where it sits in the pipeline

MiniMax H3 generates the soundtrack jointly with the video - one omni-modal model, no separate audio model bolted on afterward. That's the flagship feature and the reason a "just mute it at the end" approach doesn't work like it does for a normal video model. Sound is baked into the latent from the start, so the pack has to carry a WITH SOUND / MUTED decision all the way from the Director through VAE loading, decoding and finishing. The AudioGate is the final enforcement point: whatever state the audio is in when it reaches here, this node decides whether the finished clip carries it.

It's an internal helper (category VELVET VICE/MiniMax H3/Internal), which in practice means the reference workflow wires it for you. Its real audience is anyone hand-building an H3 graph or debugging a "why is my output silent / why is there sound" mystery.

Inputs and outputs

• audio (AUDIO) - the decoded native audio, typically from the Audio Decode gate.
• enabled (BOOLEAN, default true) - pass audio through (true) or drop it (false).

It emits two things: audio (AUDIO, or nothing when muted) and status (STRING) with the WITH SOUND / MUTED message you can surface in a text node to confirm what the render is about to do.

Installing and troubleshooting

Same story as the rest of the pack - clone into ComfyUI/custom_nodes/velvet-vice-minimax-h3 or install velvet-vice-minimax-h3 through ComfyUI Manager, then restart:

cd ComfyUI/custom_nodes
git clone https://github.com/Velvet-Vice/velvet-vice-minimax-h3

No extra pip packages; the pack leans on ComfyUI's native H3 support, and you supply the model files yourself.

The failure pattern people actually hit: they flip the Director's native_audio_output off expecting the video to come out silent, but some branch of the graph still decodes audio - or they flip the wrong gate and get a video-only render with a decode error upstream. Read the status strings at both gates before assuming the graph is doing what the toggle implies. If the encoder throws because there's no audio stream when one branch expects it, that's usually a mismatch between the two gates, not an encoder bug.