Guides
The B-Roll Gap: Generated, Selected, or Shot
ยท RenderBob team
Every cut eventually needs a shot that does not exist. AI can generate it or search a library for it, and stock libraries are answering with different rules. Here is how to choose.

Every edit hits the same hole: a line of narration and nothing to cut to. The older fixes were to reshoot, license stock, or reuse a shot that almost fits. Generation and library search are two further options, and they are different jobs.
Selected and generated
Runway's guide draws the useful line. AI-selected B-roll searches a library, so the catalogue is the limit. AI-generated B-roll makes a clip that did not exist before the description, so the description is the limit. Selection is fast, and the licence is a familiar document. Generation can be specific to the brand, and the legal position is newer.
What generation is taking first
Envato's 2026 decision guide argues that generation is displacing the generic stock categories: corporate offices, city skylines, nature atmospherics. Those are shots models can synthesise. Authentic, documentary, and real-event footage stays in demand, because authenticity is the product being sold. The guide's conclusion for professional work is that stock still wins on instant realism and licensing certainty, while generation wins on specificity, brand fit, and cost at scale. That conclusion is the vendor's.
Libraries are not on one policy
A 2026 roundup reported that Getty Images and iStock do not accept contributor AI content, that Shutterstock rejects contributor AI submissions while selling images from its own generator and running a fund for contributors whose work was used in training, and that Adobe Stock accepts generative content if it is labelled. Some libraries now mix real and AI clips, each cleared for commercial use. These policies change. Confirm the current terms with the library before you rely on a clip.
A decision rule
- Needs to be true, a real place, event, or person? Shoot it, or license footage that was shot.
- Generic atmosphere, texture, or abstraction? Generated or selected both work. Pick by speed and budget.
- An unusual combination no library has? Generate it, then review the clip.
- Something a viewer could take as documentation? Disclose it, or do not use it.
Before it ships
Production guides say to inspect generated clips for garbled text, warped logos, distorted bodies, impossible interactions, and product behaviour that is wrong. YouTube requires disclosure for realistic synthetic content that could mislead a viewer about a real person, place, or event. Filling the gap now takes minutes, so the decision is whether the shot you insert is something that happened. Keep the prompt, or the licence, next to the clip.
More from the blog
- A Risk Ladder for AI in Documentary
Screenweaver's 8 October guide ranks five documentary uses of AI by risk, from archive restoration to a synthetic face, and pairs them with EU disclosure rules now in force.
- A Character Bible for AI Filmmaking: Keeping a Cast Distinct Across Shots
Keeping one generated character consistent is hard. Keeping three distinct in the same frame is harder. A character bible has to solve both before the scenes get built.