MiniMax H3 Studio Shot Select / 镜头选择 (Advanced)
Pick One Shot Out of Your Timeline and Render Just That
- timeline
- prompt_packet
- compiled_prompt
- negative_prompt
- length
- seed
- shot_json
- report_json
Your Studio Timeline is the plan; this node is the shot list reader. MiniMaxH3StudioShotSelectT8Advanced takes a compiled timeline, lets you point at one shot by index, and hands you everything needed to render just that shot - the compiled prompt, negative prompt, duration, seed and the raw shot JSON - as plain outputs your existing Conditioning and Sampler nodes can consume. It doesn't queue anything, doesn't load models, doesn't render. It's a dumb, precise selector, and that's the point: you drive rendering, it drives data.
Why you'd want it
Rendering a whole timeline in one pass is inflexible. With this node you iterate shot by shot - render shot 3, look at it, fix its prompt, re-render shot 3 - while the timeline stays the source of truth. The shot_index input (0-based) is all you change to move between shots, and because the seed comes from the timeline's deterministic seed policy, re-selecting the same shot gives you the same seed, so your comparisons stay apples-to-apples.
The outputs, and where they go
compiled_promptandnegative_prompt- straight into your text encoder / conditioning.length- the shot's duration in frames; feed it to whatever sizes your latent or conditions.seed- into the sampler's seed input (via a converted input or a primitive).prompt_packet- the pack's structured object for the richer prompt-relay/conditioning chain, if you're using one.shot_jsonandreport_json- the raw record and the run report, for debugging and scripting.
What it won't do
No queueing, no model loading, no file writing - the description is explicit that it selects and exposes, full stop. If your workflow "does nothing" when you run it, that's expected: it's an upstream data node, and the work happens downstream in the sampler.
Installing:
part of MiniMax H3 Audio T8. Manager → search "MiniMax H3 Audio T8", or:
cd ComfyUI/custom_nodes
git clone https://github.com/T8mars/comfyui-minimax-h3-audio-T8
Restart after. No mandatory pip deps; weights are yours to supply.
The gotcha
shot_index is only meaningful against a matching timeline. Connect a timeline from a different project (or with different shots) and the node happily selects a different shot than you think - the shot_json output is how you verify which one you actually grabbed. Read it before you burn a render on the wrong shot. And as with everything in this pack, H3's local weights are license-restricted in the US/EU/UK/Korea - check your right to run them first.
Inputs (2)
| Name | Type | Default | Description |
|---|---|---|---|
| timeline | H3_T8_STUDIO_TIMELINE | — | |
| shot_index | INT | 00–100000 | — |
Outputs (7)
| Name | Type | Description |
|---|---|---|
| prompt_packet | H3_T8_PROMPT_PACKET | — |
| compiled_prompt | STRING | — |
| negative_prompt | STRING | — |
| length | INT | — |
| seed | INT | — |
| shot_json | STRING | — |
| report_json | STRING | — |