B-roll is any footage that plays over your voice while you are not on screen: a shot of the thing you are describing, a screen recording, a photo, a chart. In a talking-head short it is the main way the picture changes. This guide is editor-agnostic. The steps are the same in CapCut, Premiere, DaVinci Resolve, Descript or anything else with two video tracks.
What b-roll is for in a talking-head video
B-roll does three jobs: it shows what you are describing, it gives the viewer a change of picture, and it hides cuts. Each clip should do at least one of them on purpose.
- Show, do not only tell. If you say the kitchen was gutted, a two-second shot of the gutted kitchen does more than the sentence. This is the best use, and it is why your own footage beats stock.
- Reset attention. A single unchanging frame is easy to scroll past. A cutaway is a visual change that costs the viewer nothing.
- Cover an edit. When you remove a stumble, the picture jumps. A clip over the join hides the jump while the audio runs on.
B-roll is not decoration. A generic clip of people shaking hands over the word partnership adds a picture and no information. If you cannot say what a clip shows that your face does not, leave your face on screen. Cutaways are one of several ways to hold attention, and how to make talking-head videos more engaging covers the others.
How often to cut away
A workable starting point for a short is one cutaway every 6 to 10 seconds, each lasting two to four seconds. That is a rule of thumb, not a platform rule, and no platform publishes a required number.
For scale, the medium b-roll setting in RealTake places about six clips in a short, roughly one every eight seconds. A 45-second video at that rate spends perhaps a third of its time on b-roll and the rest on you. People came for the person, so the face should stay the main picture.
Adjust for content. A walkthrough of a process can carry more, because there is something to show at every step. A personal story or an opinion needs less, because expression is the content. Vary the gaps too. A cutaway on a fixed beat starts to feel like a slideshow.
Where to place b-roll relative to the spoken line
Place each clip so the picture arrives on the word it illustrates, or a few frames before it, and leaves at the end of the phrase. Timing matters more than the clip itself.
- Land on the noun or the verb. If the line is "we replaced the whole roof in two days", the clip starts on "roof", not on "we".
- Stay on your face for the first line. The opening seconds are where a viewer decides who is talking and whether to trust them. Open on yourself unless the b-roll is the hook.
- Stay on your face for the point and the ask. The line you most want remembered and the closing call to action land better with eye contact.
- Cover edits, not sentences. If a clip is hiding a jump cut, centre it on the join.
- Mind the captions. Stock clips often have their own text or busy detail in the lower third. Check that your captions stay readable over each clip, and avoid clips with visible words.
- Reframe for vertical. Most stock is horizontal. When you crop to 9:16, make sure the subject is still in frame.
Where to get b-roll: your own footage, stock, generated
There are three sources, and your own footage is the best of them when you have it.
Your own footage
Clips of your real work, place, product or process are specific in a way stock cannot be, and nobody else has them. Build a library on purpose. Each week, film ten short clips on your phone: your hands doing the work, the room, the tools, the screen, the before and after. Hold each shot for five seconds, shoot vertical, and name the files by what is in them so you can find them later.
Stock footage
Stock libraries cover general subjects well: cities, offices, nature, money, technology. Free libraries such as Pexels and paid ones both work. Search for the concrete object in your sentence, not the abstract idea, and skip clips that look staged. The same popular clips appear in many videos, which is another reason to prefer your own.
Generated visuals
AI image and video generators can produce a visual for a line when nothing suitable exists, such as an abstract concept or a scene you could not film. They work best for clearly illustrative images. Be careful with realistic generated footage of things that did not happen, both for trust and for platform rules, covered next.
Licensing basics before you publish
You need permission to use every clip you did not film, and you should treat a video that promotes a business as commercial use. This section is general information, not legal advice. Read the license of the specific clip.
- Free stock. The Pexels license, as read on October 3, 2026, allows free use and modification without attribution. It does not allow showing identifiable people in a bad light, implying that people or brands in a clip endorse your product, or reselling unaltered copies.
- Creative Commons. The license letters matter. NC means non-commercial use only, so avoid NC material in business videos. BY requires credit. ND forbids adaptations, and editing a clip into your own video is likely to count as one.
- Paid stock. Check whether the license covers social media and advertising, and whether a clip is marked editorial use only, which rules out promotional use.
- Footage from other creators, films or news. Assume you cannot use it without permission.
- People and property in your own footage. Get consent from identifiable clients and staff before they appear.
- Generated footage. YouTube requires creators to disclose content that is meaningfully altered or generated and looks realistic, such as a realistic scene that never occurred. Its help page says clearly unrealistic content and production help, such as generated captions, do not need disclosure.
How to add b-roll manually, step by step
The manual method is seven steps and works in any editor with at least two video tracks. Finish your cuts first. If you add b-roll and then trim the talking track, every clip slides out of place.
- Lock the talking track. Remove retakes and pauses until the audio is final. How to edit reels faster covers that step.
- Read the transcript and mark the lines. Highlight every phrase that names something visible. Cross out the first line and the main point, which stay on your face.
- Pick four to eight marks. Choose the ones where a picture adds the most, spread across the video.
- Find one clip per mark. Search your own library first, then stock. Take the first clip that is specific and clean. Do not browse.
- Place each clip on the track above your footage. Mute the clip's own audio. Start it on the key word and trim it to two to four seconds.
- Reframe and soften. Scale and position each clip to fill the vertical frame. A slow push-in over the length of the clip keeps a static shot from feeling like a photo.
- Watch it with the sound off, then on. With sound off, check that the pictures tell a rough version of the story and captions stay readable. With sound on, check that every clip arrives on its word.
Expect this to take a while the first few times. Searching for clips is the slow part, which is the best argument for building your own library.
See b-roll placed on one of your own takes
Free plan: 5 watermarked shorts a month. No credit card.
How AI video editors automate b-roll
AI video editors automate the marking, searching and placing. They transcribe the take, decide which lines should get a visual, find or generate a clip for each, and put it on the timeline at the matching word. Several tools do this, including Submagic, Jupitrr, Captions and OpusClip on its Pro plan. They differ in where clips come from and how much you can change afterwards, and our roundup of AI video editors compares them.
Automatic b-roll has predictable failure modes. It can match a keyword instead of the meaning, choose a clip that is too generic, or cover a line that needed your face. Whatever tool you use, review every clip against the checklist above and replace the weak ones.
How RealTake handles it
RealTake is our product. Its automatic b-roll draws on three sources: licensed stock footage, your own uploaded footage, and AI-generated images and clips. Your uploads are indexed by what is in them, so when you mention something you have footage of, that clip is pulled in. Mention push-ups, and your push-up demo appears. The library grows with every upload.
There are three density levels. Generation is off by default, and you approve before anything is generated. Zooms are planned so they do not collide with b-roll. You can swap, move or remove any clip in the Studio, where stock search, generation and your uploads sit in one panel. Library storage is 2 GB on the free plan and 10, 30 or 100 GB on paid plans. RealTake works from one take, in English.
"…which is why I make everyone start with push-ups—"
Sources
Pages relied on for the licensing and platform statements in this article, read on October 3, 2026. The pacing and placement advice is editing practice, not platform guidance, and is labeled as rules of thumb.
- Pexels, License. What is and is not allowed with Pexels photos and videos.
- Creative Commons, About CC Licenses. Meaning of BY, NC, ND and SA.
- YouTube Help, Disclosing use of altered or synthetic content. What must be disclosed and what need not be.
- Vendor pages for automatic b-roll features: Submagic, Jupitrr, Captions, OpusClip.