MiniMax H3 Storyboard

One text prompt gets you one clip. A storyboard gets you a scene - multiple shots, named characters who stay consistent across them, dialogue, camera moves, and background music. This is the node for the second one. It's a structured editor that lives inside ComfyUI (the node registers a /minimax-h3/storyboard panel) and compiles your shots into the exact prompt payload the Refiner and Conditioning nodes expect. If you're building anything longer than a single moment, this is the node to start from.

How it works

The storyboard holds subjects, shots, a soundscape, a music style, and a global negative prompt. Each shot carries a duration, a prompt, camera direction, dialogue, and sound. The interesting part is the subject system: you define a subject once, give it a name and a visual definition, and then reference that name in any shot's prompt. From the README's example:

Name: Alice
Definition: a woman in the appearance of <Picture 1>.

Once defined, the name works directly in shot prompts. The backend converts subject names into the <Subject N> labels H3's pipeline understands - unless the text is inside <d>...</d> markers, which are protected and passed through untouched.

Dialogue has its own convention: speaker IDs are assigned in vocal order. The README's example shows two subjects trading lines:

[Shot 1] <Subject 2> (S1) says: <d>[English] Hi Bob.</d>
[Shot 2] <Subject 1> (S2) says: <d>[English] Hello Alice.</d>

Refiners - official and third-party - normally write these IDs for you, but the manual labels are accepted too, which is useful if you want deterministic speaker assignment.

There's also a music field (music_style) that maps to non_diegetic_music in the prompt - the background score only the audience hears. The README's guidance is worth quoting: describe instrumentation ("acoustic guitar"), tempo ("slow, sparse"), and dynamic development ("gradually fading"). Do not put dialogue, singing, or sound effects there - those belong in the shot's own sound field. Leave it empty and it becomes N/A.

The inputs and outputs

No wired inputs - everything is edited in the panel. One output: prompt, a MINIMAX_H3_PROMPT built from your shots, feeding either a Refiner (the recommended path for polish) or Conditioning directly.

Installing it

The pack-wide routine - ComfyUI Manager search "MiniMax H3", or:

cd ComfyUI/custom_nodes
git clone https://github.com/xiaolibai-sys/ComfyUI-MiniMaxH3
pip install -r requirements.txt

then restart. Pure editor node - no models involved.

Common issues

• Subjects not resolving in shot prompts. Names are converted only outside <d>...</d> tags. Text inside those markers is deliberately left alone.
• Music field leaking into dialogue. You put lyrics or SFX in music_style. The field is strictly non-diegetic background; leave it empty for N/A if you're unsure.
• Storyboard output doesn't match a refiner's expectations. The refiner path exists precisely to polish this payload - Storyboard → Refiner → Conditioning is the README's own recommended chain for multi-shot work.