MiniMax H3 Source Timeline

In the pack's 0.5 architecture, Source Timeline is the front door for all your source media. Instead of wiring the same video and audio into a half-dozen nodes and hoping every consumer agrees on timing, you register the media once here and it travels as one typed descriptor (H3_SOURCE_TIMELINE) that Plan, Run Manager, previews, motion references, Loop Start, and Assemble all share. The trick that makes it cheap: video and path-backed audio stay lazy, and consumers only decode the 24 fps scene window they actually need.

What it does

Point it at media three ways:

• video_path - a direct path to a video file.
• audio_path - an optional external master-audio path; leave blank to use embedded audio or source_audio.
• source_video - a native file-backed VIDEO from core Load Video (preferred when the same asset must also be archived; the full frame tensor is never read).
• source_audio - an already-aligned external AUDIO, retained as a deferred source until Loop Start materializes its recovery copy.
• embedded_audio - auto uses one embedded stream when present, required errors when absent, ignore registers video only. External audio takes precedence.
• skip_first_frames - sets the shared timeline origin in native video frames. Skipped media stays on disk, never decoded.

Behind the single wire it records: the file-backed video descriptor, how audio is provided, native FPS and PTS origin, duration and the derived 24 fps extent, the skip offset, independent video/audio/timeline fingerprints, and original/archived/run-owned recovery locations.

Outputs are the source_timeline descriptor and a status summary of the resolved media route, timing extent, audio availability, and content fingerprint.

One deliberate scope note: this node does not choose whether the audio is final output, a reference, or generated continuity - that's audio-policy territory (Chain Policy). It just carries the media.

Where it fits

New workflows register their motion reference and any footage with Source Timeline rather than the legacy Lazy Motion AV Loader. The payoff is recoverability: the descriptor carries fingerprints and recovery paths, so a run can be resumed or assembled from disk without re-decoding the whole source.

Install and gotchas

cd ComfyUI/custom_nodes
git clone https://github.com/ethanfel/ComfyUI-MiniMaxH3-Contex-Loop.git

Restart, or install via ComfyUI Manager under "MiniMax H3 Contex Loop". No pip dependencies; a current ComfyUI with native Add Guide (PR #15439) is expected, and ffmpeg on PATH is preferred for decode. Models aren't bundled - mind the MiniMax H3 Community License territory restriction.

The main trap: paths are absolute, so shared workflows need re-pointing per machine. And if you set embedded_audio to required on a clip with no track, it errors on purpose - that's the guardrail working.