How to Generate AI B-Roll in 10 Minutes (our framework, step by step)

How We Generate AI B-Roll in 10 Minutes — No Shoots, No Videographers

So we built a system that removes it entirely. This is the exact process we use internally to manage over $107M in ad spend for our clients on Meta.

Want to see what top-performing creatives look like right now? Browse the AdLibrary Explore Tab — a daily-refreshed feed of the best-performing brands by niche.

What You Need

Just two tools:

• Gemini 2.5 Pro — generates your image prompts
• Higgsfield AI — generates the images and converts them to video

Everything runs through these two. Nothing else required.

The Two Situations

You'll always be in one of two scenarios when you need b-roll:

1. You have a shooting list prepared in advance
2. You have an idea mid-edit and need the clip immediately

The process is slightly different for each.

Situation 1: You Have a Shooting List

This is the planned workflow. You know what shots you need before you start editing.

Step-by-step:

1. Get your shooting list as a PDF
2. Upload it to Gemini 2.5 Pro
3. Paste this exact prompt:

I want you to analyze the attached file. It contains a shot list of b-roll footage I need to shoot.

Generate an image prompt for the opening frame for each shot, to be used later. You should have a clear idea of what the final video will look like. Image prompts must be in English. If possible, they should look iPhone-generated.

Gemini returns a full sheet of image prompts — one per shot.

1. Open Higgsfield AI
2. Select Soul as the AI model
3. Set ratio to 9:16
4. Set batch size to 1
5. Paste one prompt and generate

Rate the image. If the opening frame looks right, move to video:

1. Click the image output
2. Select Video
3. Paste the same prompt into the prompt tab
4. Select Kling 2.1 as the AI model
5. Set duration to 5 seconds
6. Set resolution to 1080p
7. Generate

While that renders, go back and start the image step for your next shot. By the time the next image is ready, your first video clip is done. The pipeline runs in parallel.

Situation 2: You Have an Idea On the Spot

This happens mid-edit. You need a specific clip you didn't plan for. There are two sub-scenarios.

If you have a reference image:

Download it, open Gemini 2.5 Pro, upload the image, and paste this prompt:

I must create an iPhone-generated b-roll starting from the attached image. Generate a prompt for the opening frame to be used as the starting frame. You should clearly visualize what the resulting video will look like. Image prompts must be in English.

If you only have an idea:

Open Gemini 2.5 Pro and paste:

I must create an iPhone-generated b-roll for [your idea]. Generate a prompt for the opening frame to be used as the starting frame. You should clearly visualize what the resulting video will look like. Image prompts must be in English.

From here, the process is identical to Situation 1:

• Higgsfield AI → Soul model → 9:16 → Batch size 1
• Generate the image → Rate it → Convert to video using Kling 2.1 → 5 seconds → 1080p

Total time from idea to clip: 10–15 minutes.

Why These Specific Settings Matter

Soul for the image. Kling 2.1 for the video. These are not interchangeable.

• Soul generates images that look natural and iPhone-style. The goal is b-roll that feels organic, not AI-generated. The moment a viewer clocks AI imagery in an ad, you lose them.
• Kling 2.1 handles motion better than other models for this use case.
• 5 seconds is the right duration — long enough to cut into an edit, short enough to stay sharp throughout.

If the frame isn't right, regenerate with a more specific prompt. Iterating is faster than generating multiples every time.

Studying what kinds of visuals are winning right now? Search Meta ads on AdLibrary to see how top brands are using b-roll in their creatives.

What Makes a Good Opening Frame?

A good opening frame looks like something you'd actually capture on an iPhone:

• Clear environment with natural lighting
• No text — AI almost always renders text incorrectly, so keep it out of the frame entirely
• Product check — if a product appears in the shot, inspect it carefully. AI handles objects and environments better than labels and text.

If something looks off, add specifics to the prompt before generating the video. As with everything in AI: the more precise the description, the cleaner the result.

Bonus: Swapping the Product in Existing B-Roll

Sometimes you have footage that works visually but has the wrong product in frame.

Upload the image to Gemini and prompt it to generate a modified description with the new product in place. Feed that into Higgsfield. Same workflow.

When swapping a product, specify:

• Exact product color
• Material and texture (glossy glass, matte plastic, etc.)
• Size relative to the hand or environment
• Any label text that needs to appear

Everything else — the person, the pose, the background, the lighting — stays the same.

The Real Impact

A standard b-roll shoot takes hours. Coordination, location, equipment, then editing raw footage down to usable clips.

This takes 10–15 minutes per clip.

For brands producing 20+ unique creatives per week, that difference is enormous. B-roll stops being a production bottleneck and becomes a prompt.

See what 20+ creatives a week looks like in practice. Browse the AdLibrary Explore Tab to find top-performing brands in your niche and reverse-engineer their creative strategy.

This is part of a broader AI system we use to produce full ad creatives end-to-end — A-roll, B-roll, audio, mixing existing ads with AI, editing, and identifying winners. More breakdowns coming.