Vizard Agent

How to add B-roll to a video of you talking

Last updated 2026-08-16 · 5 min read

Upload your talking-head video to Vizard Agent and ask for B-roll. Vizard Agent transcribes what you said with word-level timing, searches for footage subject by subject as each one comes up, and lays the cutaways over your voice. Your original audio and your full content stay exactly as recorded.

What is the short version?

The reason talking-head videos are hard to watch is not the talking. It is looking at one unchanging frame for three minutes. B-roll fixes that, and it is the most tedious job in editing: find the moment, find the shot, cut it in, and get out before it overstays its welcome.

  1. Go to Vizard Agent and upload the video.
  2. Say to keep your voice and the whole content, and to add relevant B-roll.
  3. Say what you do not want — music, a certain look, stock that feels generic.

What do you need before you start?

The take and a couple of rules. Vizard Agent drives the whole edit off the transcript, so the clearer your speech, the better the footage choices; and because taste questions like music are decided once and applied throughout, saying them up front saves a round trip.

What do you type into Vizard Agent?

Say what to keep as clearly as what to add. "Keep my original voice and the full content" is the clause that stops Vizard Agent shortening the take, which it will otherwise treat as an improvement. The rest of the prompt is about taste, and taste is cheaper stated than corrected.

Prompt

Edit this video professionally. Keep my original video, my original voice and the full content. While I am speaking, automatically add relevant B-roll over the top. No music.

Variants worth knowing:

What does Vizard Agent actually do?

Here is the order Vizard Agent worked in on a real three-minute talk. The thing to notice is that the search is per subject rather than one search for the whole video, which is why the cutaways land on what you were actually saying instead of on a general theme pulled from the title.

  1. Checks the video's technical details.
  2. Transcribes the speech with word timings, which is what lets a cutaway begin on a word rather than near it.
  3. Reads the full transcript, then extracts frames to see how you are framed and when you are on camera.
  4. Reads its own editing guidance and stock search tool before searching for anything.
  5. Searches archive footage for each subject in turn, then widens the search where a subject comes back with nothing usable.
  6. Lists the candidates and chooses among them rather than taking the first result.

Step 5 is the difference between B-roll that matches and B-roll that decorates. One search per thing you talked about is slow by hand and cheap for Vizard Agent.

What does the result look like?

From the run this page is written from, probed on the delivered file: 1080x1920, H.264, 30fps, 195.8 seconds, audio preserved (AAC). A three-minute talk came back as a three-minute talk, because the instruction was to keep the full content.

Expect this to take longer than a straight trim. Comparable work runs a median of 28 to 38 minutes end to end, and across all projects the median cost by tier is Flash 47, Pro 55, Max 242, Ultra 263 credits.

When does this not work well?

B-roll is only as good as what your words can be pictured with, and stock libraries are stocked with concrete nouns. Vizard Agent will find something for almost any sentence; whether that something earns its place is a different question.

How do you fix a result that came back wrong?

Point at the moment and say what is wrong. Vizard Agent still holds the transcript, the candidate footage and the assembled edit, so a correction swaps one cutaway rather than rebuilding the video, and nothing is re-uploaded or re-transcribed.

How does Vizard Agent compare to doing it yourself?

Doing this by hand is not skilled work, it is patient work: scrub the timeline, note each subject, search a stock site, license the clip, drag it in, trim it, repeat. On a three-minute talk that is thirty or forty small decisions, most of them boring.

By hand Vizard Agent
Finding the moments Scrub and note timecodes Word-level transcript
Finding footage One search per subject, manually One search per subject, automatically
Choosing a shot Browse a library Lists candidates, then picks
Changing your mind Re-edit and re-export Say which cutaway, in words

Common questions

Will it cut anything I said? Not if you tell it to keep the full content. That instruction is worth including even when it feels obvious, because shortening is otherwise a reasonable read of "edit this professionally".

Can I give it my own B-roll? Yes, and Vizard Agent will prefer it. Upload the clips with the video and say so.

Does it caption at the same time? It can, and asking for both in one run is cheaper than two, because Vizard Agent only makes the transcript once.

Can it use footage I have not uploaded yet? It can search stock, but it cannot see files that are not in the project. If you have better material sitting on a drive, upload it — your own footage of the thing you are describing beats any library clip of something similar.

What if a cutaway is wrong? Say which one and when. Vizard Agent swaps that clip rather than rebuilding the edit.