How to find B-roll that actually matches your script
The reason stock footage search feels useless is a mismatch: you search meaning, libraries index objects. The fix is to translate each script line into the physical things a camera could point at.
Abstract in, concrete out
| Script line | Weak search | Search that works |
|---|---|---|
| The company had been quietly insolvent for two years. | insolvency, bankruptcy | empty office desks, unopened mail pile, dark boardroom |
| Investors kept sending money anyway. | investment, trust | phone banking app transfer, handshake in lobby, wire transfer screen |
| By spring, the fund had collapsed. | economic collapse | closed shutters storefront, falling stock ticker, empty trading floor |
Rules for visual keywords
- Name a subject a camera can see: person, object, place, action.
- Two to four words. Long phrases collapse most stock library search relevance.
- Add a modifier only if it matters: night, aerial, close up, slow motion.
- Avoid named brands, real people and dated events — stock libraries rarely have them.
Keep the sequence consistent
Once you have candidates, check them as a set rather than individually. Mixed colour temperature, mismatched grain and jumping time of day are what make an otherwise good selection feel stitched together. Prefer a slightly weaker clip that matches the look of its neighbours.
Check the licence before you cut
Free libraries such as Pexels and Pixabay allow commercial use under their own licences, but restrictions on identifiable people, logos and redistribution still apply. Save the source URL for every clip you use so the licence is traceable later.
Common questions
- What is B-roll?
- B-roll is the supporting footage cut over narration or an interview: locations, hands, objects, crowds, archival material. It carries the story visually while the voice carries the information.
- Why does stock footage search return nothing useful?
- Most stock libraries match tags on the clip, not the meaning of your sentence. Searching an abstract phrase like 'economic collapse' returns charts and stock photos; searching the concrete objects in that idea — empty shop, closing shutters, queue at a bank — returns usable footage.
- How many clips do I need per minute of narration?
- At an average shot length of 4–8 seconds, one minute of narration usually needs 8 to 15 distinct clips. Collect more candidates than you need so you can avoid repeating a look.
Skip the translation step
Overdrawn reads each scene, writes the concrete visual concepts for you, and returns matching clips from Pexels and Pixabay in your chosen aspect ratio.
Try it on your script