A director’s console for MiniMax H3 long-form video

What it does ComfyUI_MiniMaxH3_Director is a single, sprawling ComfyUI node that acts as a director’s console for the official MiniMax H3 video generator. It bundles timeline editing, scene splitting, conditional encoding, sampling, and audio-video export into one interface, letting you chain segments with different modes—text-to-video, image-to-video, first-last-frame, reference-based, and video-to-video—without rewiring the graph between shots. Under the hood it still calls the official MiniMaxH3ImageToVideo and related nodes; this is essentially a very ambitious orchestration layer with a built-in timeline UI.

The interesting bit The standout trick is segment guidance: the node can pin the last N frames (and any generated audio) from one finished segment into the next segment’s sampling context, then crop the overlap so motion flows across cuts instead of restarting cold. It also supports a full refine and upscale pass—including H3 latent upscaling and optional RTX VSR—after the initial generation, and lets you run only selected segments to save GPU time.

Key highlights

• Timeline-based segmentation with manual cuts, even splits, or PySceneDetect auto-split, plus thumbnail previews per segment
• Six task modes in one node: t2v, i2v, fl2v (first-last frame), r2v (reference subject), v2v, and rv2v
• Segment-to-segment motion handoff via configurable context frames (default 22) to bridge cuts
• Native stereo audio generated alongside video; v2v and rv2v can generate new sound, keep the original track, or mute
• External group wiring via Director Group nodes so upstream preprocessing can feed the timeline in batches
• Optional post-generation refine/upscale with refine, upscale, or latent_upscale modes and multi-pass support

Caveats

• Requires ComfyUI v0.30.0+ and the official MiniMax H3 nodes already installed; this is an add-on, not a standalone pipeline.
• The primary documentation and UI are Chinese-first, so non-Chinese speakers will need to rely on the secondary English README.
• If video decoding fails, the node outputs gray placeholders rather than generated frames, which the README notes explicitly rather than hiding the failure.

Verdict If you are already generating with MiniMax H3 inside ComfyUI and need videos longer than a single clip, this node is basically mandatory; for one-off short clips, the official nodes are less overhead.