MiniMax H3 Voice Context Review Gate / 首段人工确认门 (Advanced EXP/T8)

The Voice Context node compiles a careful plan for keeping the right voice on the right character across long H3 segments. But a plan is only as good as the first shot that tests it - and if you've ever let an automated chain run for an hour only to discover segment one has the wrong voice, you know why that first shot deserves a checkpoint. This node is that checkpoint: a small planning gate that holds the compiled segment prompts until a human flips the approval switch.

The honest framing matters here, because this node's name invites a misunderstanding. It is not a pause between segments - the pack is explicit that a real per-segment pause is the Background/Accepted workflow's job. This is a planning gate: it sits between the Voice Context compiler and the long-video runtime, and it controls the release of the JSON, not the rendering. If you want to literally stop the machine and listen after every segment, you want the other workflow.

How to use it

Two inputs, two jobs:

• voice_context_plan - from the Long Video Voice Context node
• first_shot_approved - BOOLEAN, defaults to false. Leave it false until you've rendered and checked the first shot's voice; flip it true to let the prompts flow

Outputs: segment_prompts_json and audio_pin_frames_json (the release - pass these to the runtime), released (BOOLEAN - tells you whether the gate is open), and report_json. Wire released into whatever downstream logic decides whether the chain can proceed, and it doubles as your "did I actually approve this?" status.

The practical rhythm the tooltips and docs imply: compile the voice plan, feed this gate with first_shot_approved=false, render a test first shot, audition the voice, then set first_shot_approved=true and let the rest run. If the voice is wrong, you fix the bindings or timeline in the Voice Context node and recompile - the gate has saved you from committing that mistake across every segment.

Installing it

Part of the T8mars/comfyui-minimax-h3-audio-T8 pack - search "MiniMax H3 Audio T8" in ComfyUI Manager, install, restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/comfyui-minimax-h3-audio-T8.git minimax-h3-audio-T8

Update ComfyUI itself first (recent core: comfy_api.latest, comfy.weight_adapter, comfy.patcher_extension, comfy.ldm.minimax); requirements.txt is intentionally empty.

Where people get burned

Two ways. First, assuming the gate pauses rendering - it doesn't; if you connect it expecting a human-in-the-loop stop after every segment, you'll be surprised when the chain keeps going. Use the Background/Accepted route for that. Second, flipping first_shot_approved true before you've actually listened to a first shot, which is the default-false design trying to save you from yourself. One more subtlety: the gate releases what the plan compiled - it doesn't re-validate the plan's contents. If the voice bindings were wrong upstream, approval just blesses a wrong plan into the runtime. The gate is a process checkpoint, not a quality audit; do your plan-checking in the Voice Context node and your human-checking here.