The Short Answer
To make AI b-roll that a client actually uses, watch their existing footage before you generate anything and match its colour, lighting, and lens feel. Then produce single-movement clips a few seconds longer than the gap they need to fill, with no text, no logos, and no faces. B-roll fails on mismatch far more often than it fails on quality.
Why AI B-Roll Gets Rejected
The clip that gets cut is rarely the bad-looking one. It is the one that is beautiful and wrong. Warmer than the rest of the video, shot on a different apparent lens, graded in a way that pulls the eye, or moving at a speed that does not match the pace of the edit around it.
An editor drops a clip into a timeline, sees it stand out, and removes it in about two seconds. They will not tell you it was a colour mismatch. They will say it did not fit, and you will spend the next round improving the wrong thing.
Watch the Footage First
Before generating anything, ask for five minutes of the actual video, or the previous one they made. Then note the things that decide whether your clip will blend, because those notes become the constant part of every prompt you write for that client.
- Warm or cool overall, and how strongly.
- How much depth of field. Heavily blurred backgrounds or mostly sharp.
- Handheld with slight movement, or locked off and still.
- Cutting pace. Whether shots hold for four seconds or one.
- Contrast. Whether the blacks are crushed or lifted and flat.
- Apparent focal length. Wide and slightly distorted, or compressed and telephoto.
Those six notes go into every prompt for that project. It is repetitive and it is the entire reason your clips will survive the edit while somebody else's get deleted.
What Not to Put in the Frame
Some content is unusable in client work regardless of how good the generation looks, and generating it anyway is how you produce a folder of clips that all get rejected for the same reason.
Avoid, and what to do instead
| Avoid | Why | Instead |
|---|---|---|
| Any readable text or signage | Generated text is usually wrong and it draws the eye immediately | Frame so text is out of focus or out of frame |
| Logos or branded products | It is not the client's brand and it creates a rights question | Generic unbranded objects |
| Recognisable faces looking at camera | Raises consent questions and pulls focus from the narration | Backs of heads, hands, or figures at a distance |
| Screens showing interfaces | Generated interfaces look wrong to anyone who uses software | Screens angled away, glowing but unreadable |
| Tight close-ups of hands doing a task | The most common place for anatomical failure | Wider framing where hands are part of the scene |
Generate for the Timeline, Not the Preview
Editors need handles. If a gap in the edit is four seconds, they want six or seven so they can slide the clip around and choose the in and out points. Delivering exactly four seconds means any adjustment forces them back to you.
- Generate a few seconds longer than the gap you were told about.
- One continuous movement per clip, with no internal cut or change of direction.
- Deliver at the highest resolution available, since it is easier to scale down than to recover detail.
- Give three variations of any shot they described specifically, because their mental image will not match your first attempt.
- Name files by their role in the edit, such as opener-city-wide-01, not by generation number.
How to Hand It Over
Send a single folder with clearly named files and a short note that says which clip you think goes where and why. That note is what turns a delivery into a service. It takes three minutes and it is the difference between being the person who sends files and the person who understands the edit.
Be direct about anything you could not do well. If they asked for a shot of someone typing and you know close-up hands will look wrong, say so and offer the wider alternative you generated instead. Naming a limitation before they find it builds far more trust than quietly delivering the weak version.
Why This Is Good Work to Sell
B-roll is a defined, repeatable job with an obvious before and after. The client already knows they need it, already has gaps in an edit, and can judge the result in seconds. That combination makes it far easier to sell than an open-ended offer to make videos with AI.
It also compounds. Once you have matched a client's look, every future project for them is faster, because the six notes are already written and you already know which shots survive their edit. The second project is where the margin is.
Where to Go From Here
Take any video already published by a business you would like to work with, write the six matching notes from it, and generate three clips that could drop into it unnoticed. That set is both your practice and the most persuasive thing you could put in front of them.
Inside Higgsfield Income Club, members post the client footage alongside the b-roll they generated for it, so you can see which matches held up in a real edit. Join at higgsfieldincomeclub.com for $9/month.
Short, practical drops on income paths, prompts that work, packaging for views, and getting paid. No spam, unsubscribe anytime.
Frequently asked questions
Why does my AI b-roll not fit the client's video?
Almost always a match problem rather than a quality problem. Colour temperature, depth of field, contrast, and apparent focal length all have to line up with the surrounding footage. A beautiful clip that stands out gets deleted in about two seconds.
How long should each b-roll clip be?
A few seconds longer than the gap it fills, so the editor has handles to slide the clip and choose in and out points. Delivering the exact length means any adjustment sends them back to you.
What should never appear in AI b-roll for clients?
Readable text, logos, branded products, recognisable faces looking at camera, and readable screen interfaces. All of those either generate incorrectly or create a rights question you do not want in a client deliverable.
How many versions of each shot should I deliver?
Three for anything the client described specifically, because their mental image will not match your first attempt. Generation is cheap and choosing from options feels like service rather than a take-it-or-leave-it approval.
Should I tell the client what I could not do well?
Yes, and before they find it. If close-up hands will look wrong, say so and offer the wider alternative you generated instead. Naming a limitation up front builds more trust than quietly shipping the weak version.
What should I watch for before generating anything?
Five minutes of their existing footage. Note whether it is warm or cool, how blurred the backgrounds are, whether it is handheld or locked off, the cutting pace, the contrast, and the apparent lens. Those six notes go into every prompt for that client.
Last reviewed by David on August 15, 2026
