AI B-Roll Generator for Ads: Shot List and Prompts

An AI B-roll generator makes the short supporting shots in an ad, such as product close-ups, hands, textures and lifestyle scenes, from a text prompt or a product photo instead of a shoot. For ads, the reliable method is image-to-video: create one still per shot with your real product in it, animate each still for about five seconds, and keep only the cleanest two to three seconds.

BuzzVideo AI does not have a button called B-roll. You make it with the image and video tools inside the Marketing Agent, which is how we produced the four-shot example in this post from a single product photo. Below are the shot list, the prompts, what we had to cut, and what it cost.

Try it: Upload one product photo and ask for a 9:16 macro start frame, then animate it for five seconds. Open the Marketing Agent →

What is AI B-roll, and what is it for in an ad?

B-roll is everything that is not the main talking or selling shot. In a 15-second product ad it does three jobs: it stops the scroll with something physical in the first second, it gives the eye a new picture every two to three seconds, and it covers cuts in a voiceover or a creator clip.

Tools sold as AI B-roll generators fall into two groups, and they solve different problems:

Kapwing's B-roll generator is an example of the second group and HeyGen's B-roll tool of the first, going by their own product pages. If your video is a long talking-head piece, a transcript-matching tool will be faster than what we describe here. If the ad has to show your product, generate from your own photo.

Which B-roll shots does a product ad need?

Four shots cover most short ads. Each has one job, one subject and one movement.

That gives you 8 to 12 seconds of footage, enough for one ad or for the middle of a creator video.

How do you make AI B-roll from one product photo?

Step 1: Start from a photo where the label is readable

Use a straight, well-lit product photo. Every later shot inherits the label from this image, so a blurry or angled label will only get worse.

Step 2: Generate one start frame per shot

In the Marketing Agent, open Generate Image, attach the product photo as the reference, choose 9:16 and describe one shot at a time. We used GPT Image 2.5 Sunburst. Check each still at full size before animating it: label spelling, one product only, the right number of fingers.

Step 3: Animate each start frame for five seconds

Open Generate Video, attach the start frame, choose 9:16 and five seconds, and turn sound off if the model allows it. We used Kling 3.0 at 720p. Describe one camera move and one thing that changes, and say what must stay still.

Step 4: Inspect, trim, and cut together

Step through every clip at half-second intervals, keep the cleanest two to three seconds of each, and join them with hard cuts.

Copy-paste prompts for the four shots

These are our prompts with the product details replaced by brackets. The start-frame prompts go to the image model with your product photo attached; the motion prompts go to the video model with the start frame attached.

Shot 1: hero macro

“Vertical 9:16 photorealistic macro product photo. The exact [product] from the reference image, same shape and label text, stands upright in the centre on [surface]. Warm low sunlight from the left, solid [colour] background, shallow depth of field. One product only. No hands, no people, no extra text, no logos.”

- Start frame

“Use the image as the exact first frame. Very slow, smooth push-in. Warm sunlight sweeps across the product from left to right. The product does not move; the label stays sharp and unchanged. No hands, no people, no new objects, no added text, no cuts.”

- Motion

Shot 2: hand

“Vertical 9:16 photorealistic lifestyle photo. A woman's left hand only (one hand, five fingers) enters from the left edge and holds the exact [product] from the reference image upright just above [shelf or table], label facing the camera. Soft-focus [room] behind. Only one hand in the image, no face, no body. One product only. No extra text, no logos.”

- Start frame

“Use the image as the exact first frame. Locked-off camera. The hand slowly sets the product down upright, label facing the camera, lets go and withdraws out of frame to the left. The product then stands still. Only one hand ever appears; the label stays unchanged. No face, no new objects, no added text, no cuts.”

- Motion

Shot 3: texture

“Vertical 9:16 photorealistic top-down macro. A clear glass dish on [surface] holds a shallow pool of [the product's texture and colour]. A [dropper, spoon or spatula] lies across the lower right with one drop hanging from its tip. Hard warm sunlight. No label, no text, no hands, no people.”

- Start frame

“Use the image as the exact first frame. Top-down, camera locked off. The single drop falls into the pool and sends one slow ripple across the surface. Nothing else moves. No hands, no people, no new objects, no text, no cuts.”

- Motion

Shot 4: lifestyle

“Vertical 9:16 photorealistic still life. The exact [product] from the reference image stands upright on [vanity, desk or counter], label facing the camera, beside [two or three simple props]. Soft morning window light, sheer curtain, leaf shadows on the wall. No mirror or other reflective surface that could show a person. One product only. No people, no hands, no extra text, no logos.”

- Start frame

“Use the image as the exact first frame. Slow sideways dolly of a few centimetres, product in focus. The curtain moves gently and the leaf shadows sway. The product and props do not move; the label stays unchanged. No people, no hands, no mirrors, no new objects, no added text, no cuts.”

- Motion

Example: a four-shot B-roll reel from one photo

Nothing here was filmed; the only input was one flat-lay photo of the bottle and its carton. The reel is deliberately bare so you can judge the footage. In a real ad you would add voiceover or creator audio, and put the offer as a text card on the lifestyle shot, where there is empty wall.

What went wrong, and what did we cut?

Five generations gave us four usable shots. Here is what the frame-by-frame check found.

The lesson from the mirror is general: a video model will try to fill any reflective surface. Take mirrors, windows facing the camera and glossy screens out of the start frame instead of asking the motion prompt to keep them empty. Even after that, the second try slid what looks like the edge of a new mirror into one corner, which we trimmed and cropped out. That is why the lifestyle prompt above now rules reflective surfaces out, and why you still check every clip.

• Label: pause on three frames per clip and read every word.
• Hands: count fingers at the start, in the middle and at the release.
• Uninvited people: check mirrors, windows and the edges of the frame.

How much does AI B-roll cost?

These are the credits our workspace was charged today for this example. Per-model costs change, so check the pricing page before you plan a batch.

On the Pro plan billed yearly ($35 a month for 4,900 credits), 425 credits is about $3. Starter ($14 a month billed yearly, 1,900 credits) includes Kling 3.0 and the image models used here; the Free plan has 500 one-time credits and no video generation. See pricing for current numbers, and our cheapest AI video generators comparison if cost per clip is your main constraint.

Do you have to label AI B-roll in ads?

Often, yes. TikTok's ad policy requires clear disclosure when AI-generated content shows realistic scenes or people, and Meta says it labels ads made or significantly edited with generative AI, including ads it detects were made with third-party tools (Meta Help Center; Meta's announcement). Product B-roll like this is realistic by design, so plan for the label instead of trying to avoid it. Our guide to labeling AI ads on Meta and TikTok covers both platforms.

Two rules of our own: do not use generated B-roll to show a result the product has not delivered, and keep prices, dates and claims out of the generated footage; add them as text in the edit.

If you have one good product photo, you can have a first set of B-roll today. Open the Marketing Agent, generate the four start frames, animate each one, and keep only what survives a frame-by-frame check. For full ads built the same way, see product demo videos from photos.

FAQs

Last verified: November 10, 2026, against our live pricing and product pages.

Next reads: Product demo videos from photos · AI product commercial from photos · Kling 4.0 for ads · Edit a video with ChatGPT or Claude