MiniMax H3 Scheduled Video Ref
Schedule one performance video per scene, soundtrack included
- video
- audio
- previous
- schedule
- schedule_fingerprint
- status
Sometimes one scene needs to move like the dance clip, another like the fight clip, and the loop shouldn't replay the same performance everywhere. This node registers one 24 fps reference video - plus its optional synchronized soundtrack - under a stable @tag, scopes it to the scenes you pick, and lets the scheduler assign compact <Video N> and <Audio N> labels per scene. It's the video counterpart to the scheduled picture ref, and it pairs cleanly into the same chain.
What it does
Connect your video (reference frames at 24 fps - if your loader source runs at another rate, run it through Reference Video Prep first). Set a tag like performance (becomes @performance in prompts; not a native number). Set scenes - blank for all, or 1, 1:4, 1,3,5:8. The audio input is the optional soundtrack of this same video; it stays index-paired with its video in stock Ref2VA and gets its own audio_tag (blank derives @<video_tag>_audio). Chain with previous for stable priority order.
Then the interesting one - timeline_mode:
- restart_each_scene (default) - every active scene receives the reference from frame 0. Simple, predictable, and what the original behavior always did.
- sequential - the 24 fps source advances along the Plan timeline, repeating the same overlap as Motion Context. This is the one to pick when your reference is one long continuous performance and each scene should see the next part of it. It requires Current Shot
stateconnected to Scheduled Ref2VA.
Outputs: schedule (chain it or feed Scheduled Ref2VA), schedule_fingerprint, and status.
Like the other scheduled nodes, this one never inserts prompt text - you write @performance in the Plan prompt and the scheduler resolves it. And the tags are aliases, not fixed native numbers: active videos renumber from <Video 1> each scene, so removing an earlier entry doesn't break your prompt.
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. Models aren't bundled - check the MiniMax H3 Community License territory restriction.
Where people get caught: sequential mode with no state connected to Scheduled Ref2VA. The scheduler can't know scene starts without it, so the mode simply won't behave. And don't feed this node a video at the wrong frame rate - it assumes 24 fps input; normalize first.
Inputs (7)
| Name | Type | Default | Description |
|---|---|---|---|
| video | IMAGE | Reference video frames at 24 fps. Use Reference Video Prep when the loader source has another frame rate. | |
| tag | STRING | performance | Stable alias such as @performance. It is NOT a native Video number. Active videos are renumbered from <Video 1> per scene, so this @tag remains valid if an earlier entry is removed or inactive. |
| scenes | STRING | Scenes where this video and its optional paired soundtrack are active. Blank means all; 1, 1:4, and 1,3,5:8 are supported. Only active videos consume <Video N> numbers. | |
| audio_tag | STRING | Alias for the paired soundtrack when audio is connected. Blank derives @<video_tag>_audio. This is also a stable alias, not a reserved <Audio N> number. | |
| timeline_mode | COMBO | restart_each_scene | restart_each_scene preserves the original behavior: every active scene receives the reference from frame 0. sequential advances the 24 fps source along the Plan timeline, repeating the same overlap as Motion Context. Sequential mode requires Current Shot state connected to Scheduled Ref2VA. |
| audioopt | AUDIO | Optional soundtrack of this same reference video. It stays index-paired with the video in stock Ref2VA and receives its own audio tag. | |
| previousopt | H3_REFERENCE_SCHEDULE | Optional preceding scheduled reference chain. It sets stable priority order, not permanent native label numbers. |
Outputs (3)
| Name | Type | Description |
|---|---|---|
| schedule | H3_REFERENCE_SCHEDULE | Reference schedule to chain into another entry or Scheduled Ref2VA. |
| schedule_fingerprint | STRING | SHA-256 of all sources, tags, and selectors for checkpoint safety. |
| status | STRING | Normalized video/audio tags, selector, entry count, and fingerprint. |