Parameter nodes — shared behaviour

Parameter nodes are pickers on the canvas. They never call a model and never produce a job: each one composes a prompt fragment that is injected into a downstream node’s prompt, either through that node’s cinematography handle or by wiring the picker straight into a Text Prompt / Combine Text / LLM Chat input.

The per-node pages in this folder list each picker’s own fields and catalog. This page covers the fields every picker shares.

Free text: preText / postText

Two optional free-text fragments composed around the structured hint:

preText, <composed hint>, postText

Both are plain strings, both default to empty, and an empty one contributes nothing (no stray commas). Use them for one-off detail the catalog does not cover — "wet-haired, covered in paint" before, "wearing a leather jacket" after.

Prompt hint mode

Every catalog-backed picker can inject its selection in one of two registers:

hintMode	What gets injected	Example (Framing → medium close-up)
full (default)	The long descriptive hint — the full clause, written for models that respond to prose.	medium close-up, subject framed from the chest up
compact	The short professional term — the words a crew would actually use.	medium close-up

Both registers describe the same selection. full gives a diffusion model more to hold on to; compact keeps the prompt short when many pickers feed one generation, or when the model already knows the trade vocabulary and the long form only dilutes it. preText / postText are appended in both modes.

Where to set it. Two places, same lever. On the canvas: hover (or select) any picker node — the control row at the top of the card carries a Prompt hint toggle on the left — Full / Compact — beside the existing Picks / Prompt / Both display-mode toggle. Switch the node to the Prompt or Both display mode to read the exact fragment the node will inject; the preview updates with the toggle. In the config panel: the same toggle sits beside the Will inject into prompt preview at the top of every picker’s panel, and that preview shows the same composed fragment.

Defaults and compatibility. The field is absent on nodes created before the lever existed, and absent means full unless the node is composed by a compact composer (see Inheritance below) — no existing workflow changes behaviour. Nothing is rewritten when you switch: the mode is stored as hintMode in the node’s data, so it travels with copy/paste and node duplication, is captured by node presets, and survives export/import.

Inheritance. Two pickers compose other pickers into one sentence: Camera Motion and Transition, through their startState / endState handles. Setting either of those to compact propagates the mode into the pickers wired to those handles, so the composed sentence stays at one level of detail instead of mixing a short term with two paragraphs. A wired picker that declares its own hintMode always wins over what it inherits — including an explicit "full", which is how you opt one input back out of a compact composer. A picker read on its own (the canvas preview, the config-panel preview, a direct wire into a Text Prompt or an LLM Chat) inherits nothing and uses its own mode, so the same node can read full in its preview and inject compact inside a compact composer.

Character FX and Character Motion compose a target too (Character Motion also a partner), but those are character / face / object / location references rather than pickers — each contributes a NAME, not a fragment, and that name is injected in both modes.

Authoring workflow JSON directly (MCP, SDK, or an imported file): set it on the picker node’s data.

{
  "id": "framing-1",
  "type": "framing",
  "data": { "label": "Framing", "shotSize": "medium-close-up", "hintMode": "compact" }
}

Only "full" and "compact" are meaningful; any other value is treated as undeclared — which means full, or the mode inherited from a compact composer. Compact is opt-in, so a typo can never silently shorten a prompt.

Which nodes have it. Every picker listed under Picker nodes in the node reference — all five families: Look, Camera, Subject / Object, Multi-dim composed pickers (Framing, Lighting, Person, Styling, Temporal, Exposure Settings) and Sound / Music / Voice. The free-text and pure-runtime parameter nodes (Tone, Style Guide, Provider, Scene Count, Duration, Aspect Ratio, Motion) have no catalog term behind them, so they carry no hint-mode toggle.