MiniMax H3 Clip
Dropping supplied footage into the reel
- reel
- audio_vae
- reel
Some shots in a piece aren't generated - they're footage you already have. MiniMax H3 Clip Reel is how that footage gets into a Creator Palette render: it adds a supplied clip to the reel without decoding it. It's the one node in the chain that never turns video into pixels.
That's a bigger deal than it sounds. The finished video is written frame by frame at mux time, so a clip only ever has to be named here - the muxer demuxes the file, conforms it, and re-encodes it straight into the container. Two minutes of 768p footage would be ~35 GB as a tensor; this way it's a tiny dict on the reel. It's an MiniMax/internal node, placed by the Timeline rather than from the menu, but it's worth understanding because it's how you mix reference clips and existing b-roll into an otherwise-generated piece.
How it works
You feed it clip_data - a JSON blob naming the file (resolved against ComfyUI's input folders) plus what role the clip plays in the render. It appends {"clip": spec} to the reel, nothing more. The one subtlety is audio: if the clip plays with its sound, the node needs audio_vae - not to decode anything, but for one number, the sample rate its decoder outputs. The generated passes' sound arrives at that rate, so the clip's sound gets resampled to match. Miss the VAE and the clip would play at the wrong pitch in the final file, which is exactly the kind of silent mistake the pack refuses to make.
Inputs and output
clip_data- the clip spec (path, role, sound on/off). Required.reel- optional, the passes and clips already in front of this one. Absent on the first element.audio_vae- optional, only for the output sample rate; required when the clip plays with its sound.
Output is a single reel - the same MMC_REEL type the Reel and Save nodes use, with this clip appended.
Where it sits in the pipeline
The Timeline writes a Clip Reel when you add a footage card to a shot list. Around it you'll see the clip's seam helpers - Clip Frames and Clip Audio - which hand before/after generations the clip's opening or closing frames and sound. A clip nothing continues from never gets decoded at all; that's the point of the seam nodes being separate from the clip's reel node.
Install is the shared pack install: ComfyUI Manager → search "H3 Creator Palette", or git clone https://github.com/z3rofeels/ComfyUI-H3-Creator-Palette into custom_nodes/, restart, hard-refresh. If you hand-build a graph with a sound-on clip but no audio_vae, the node raises and tells you exactly which input is missing - better than discovering the pitch shift in the render.
Inputs (3)
| Name | Type | Default | Description |
|---|---|---|---|
| clip_data | STRING | — | |
| reelopt | MMC_REEL | — | |
| audio_vaeopt | VAE | Only for its output sample rate — the clip's sound is resampled to whatever the generated passes decode at. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| reel | MMC_REEL | — |