From a sentence to five search queries and one generation prompt
A prompt that turns each line needing a cutaway into five queries for your own footage or a stock site, ordered from literal to loose, plus one generation prompt as the fallback when nothing is found.
For each sentence below, produce search queries to find a cutaway that shows it literally, then a generation prompt as the fallback. Per sentence: SENTENCE: quoted. THE THING TO SHOW: one noun phrase, the object or action the sentence is about. QUERIES, 5, from most literal to loosest, each 2-5 words, no adjectives like 'beautiful' or 'cinematic': 1. the exact object + the exact action 2. the object + the setting 3. the object alone, close-up 4. the action alone, hands only 5. the setting alone, wide FILTERS: orientation {9:16 or 16:9}, duration 5-15 s, no people's faces, no logos. GENERATION PROMPT (fallback, one paragraph): subject with 2 visual details, one action, setting, shot size + one camera move, light, style lock: {style lock}, ambient sound, 5 s, 'no on-screen text, no faces, no logos'. FOUND IN OWN FOOTAGE? If a clip in the list below shows the thing, name it first and skip the queries. Own footage (optional, clip + description): {paste clip list} Sentences: {paste sentences}
In Claude Code, paste the prompt as a message.
Steps
- Search your own footage first; the sidecar descriptions (or Life2Film's segment list) are what the queries run against before any stock site.
- Run queries 1-3 on a stock site; if 4 and 5 are what you end up using, the shot is no longer literal, and the generation prompt is the better fallback.
- Put whichever clip you choose on the contact sheet with its sentence; a stock clip can be off topic as easily as a generated one.
Tips
- Hands-only queries (4) find usable footage for almost any 'doing' sentence without a face to match.
- Keep the query sets with the project. The next video on the same topic reuses half of them.
| works with | Claude Code: Runs as a skill or a prompt in the terminal; can read your files and write the result. · Claude: Paste into claude.ai or the Claude app. · ChatGPT: Paste into ChatGPT; works with any GPT model. · Any LLM: Model-agnostic: Gemini, Mistral, a local model. |
| cost | Free. No API, no key. |
| status | Unverified: written from the technique, not yet run end to end by us. |
| license | CC BY 4.0 for the text here; skills are MIT. |
| updated | 2026-11-01 |
| for agents | broll-search-query-generator-prompt.md |
Related
Generated b-roll or real: the decision rule
When a generated cutaway is fine (faceless, no logos, no claim about a real place or time) and when only footage from the same day will do. Five questions, in order, before you open a video model.
Cutaway matching: sentence to a literal visual, same place and time
A prompt that reads each sentence of a transcript and names the one picture that shows it, with the place and time it must come from, so the shot list is a list of things, not moods.
Generate a 10-clip b-roll pack for a talking-head video
One agent prompt that reads your transcript and writes ten generation prompts for cutaways that show what is being said, each tagged with the sentence it covers.
B-roll shot list from a transcript: the table to film or generate from
An agent prompt that reads a transcript and returns a shot table (timestamp, line, shot, source: have / film / generate) for a camera or a video model, grouped so each location is filmed once.
The cutaway rule: what is said, same place and time, never over moving lips
Three constraints for every cutaway in a talking film, and the agent prompt that applies them to a transcript. Kills what viewers read as broken: a picture from elsewhere, a face out of sync.
A mentioned object must be seen, ideally just before the line about it
The rule that turns 'X is mentioned, never shown' from a viewer complaint into a check you run on the transcript: every named thing gets a shot, placed just before or on the line.