write_script

Generates a narration script from written context and optional footage analysis. Uses an LLM to write short-form video narration optimized for spoken delivery.

In the editor

Add it from the Library (AI group), or with + add step and a search for write_script.

Everything it needs can be typed in the Inspector. Put what the video is about into context_text: the product, the audience, the points that must be made. If that material already lives in a file you uploaded, add a context file input instead. target_duration_sec is how long the finished narration should take to say, and style is a dropdown for the tone.

Turn on suggest_ideas if you would rather choose a direction than accept the first one. The run then pauses and offers a handful of angles to pick from before the script is written, described under suggest_ideas below. Tick Review gate as well to see the finished script before the run carries on.

Nothing goes on the timeline. The script becomes a piece of text that later steps read: a voice step to speak it, a teleprompter step to show it to you while you record, or a prep step to tidy it for speaking. Wire this step into those as their source or script input, choosing step and picking it from the dropdown.

Inputs

ID	Type	Required	Description
`context`	file	no	Plain-text context files about the subject (product description, brief, etc.)
`footage_analysis`	step	no	Output of `analyze_footage`: used to match script topics to available visual moments

At least one of context_text param or a context file input must be provided.

Params

Param	Type	Default	Description
`target_duration_sec`	duration (s)	`45`	Target spoken duration in seconds (~150 wpm)
`style`	select: `promotional`, `educational`, `conversational`, `formal`, `casual`, `explainer`, `documentary`	`"promotional"`	Tone and style of the generated script
`suggest_ideas`	bool	`false`	When true, generates 3–5 script directions and pauses for user selection before writing
`context_text`	text	none	Inline context text (alternative to file input)

Output

script: the narration script text, which later steps can read with step: <this step's id>.

suggest_ideas

When suggest_ideas: true, the step works in two passes:

Ideas. It proposes 3 to 5 distinct script directions, each with a title and a short description.
Your choice. The run pauses and shows you the directions. Pick one with Choose this concept, or use Write my own direction to describe what you want instead. No preference lets it decide, ↻ Suggest again asks for a new set of ideas, and Cancel run stops there.

Whatever you choose steers the script it then writes. See Review gates.

If you are unhappy with the script afterwards, use the ↩ control on this step in the run view: it re-picks a concept from the ones already suggested, and its ⋯ menu offers Generate new concepts or Keep same concept.

Example

- id: write_script
  processor: write_script
  params:
    target_duration_sec: 30
    style: promotional
    context_text: |
      A portable espresso maker for travelers. Key features: fits in carry-on,
      no electricity required, makes barista-quality espresso in 60 seconds.
    suggest_ideas: true
  input:
    - id: footage_analysis
      step: analyze_footage
  review: true