Skip to content
FO Flash OmniIndependent guide Not affiliated with Google

Video generation · Workflow guide

AI B-Roll Generator: Cinematic Clips From Text

B-roll is a supporting shot, not a second story. The best prompts are short, specific, and built around one camera move. If the clip needs a plot, it is not b-roll.

  • Independent guide
  • No sign-in here
  • Prompt runs locally
  • Generator opens flashomni.ai
Cinematic aerial view of a city at dusk with long car light trails
Concept image generated with GPT Image 2 for this guide. It illustrates a b-roll frame, not a guaranteed output.

Quick answer

Describe a single shot with one subject, one action, one camera move, and one lighting idea. Avoid plot. Generate several five-second clips and cut them into the edit. Match light direction and movement to the surrounding footage.

At a glance

Best input

A shot list and one clear visual idea per clip

Best output

Five-second cutaways that match the edit

Video models

Veo 3.1, Sora 2, Kling 3.0, Hailuo, Gemini Omni Flash

Next step

Cut the clips to the music or voice track

Build your prompt

Pick a preset, adjust the subject and scene, then copy the prompt. The optional photo preview stays in your browser: this page does not upload it. The generator is a separate site with its own terms and pricing.

1. Choose a starting point

No file selected
Local preview of the selected photo Your photo is not uploaded by this page. It stays in this browser tab.

2. Copy the prompt

Open Flash Omni

Copying the prompt does not create an account or charge credits. The hosted generator opens a separate site with its own terms and pricing.

What you can make

City and travel aerials

Dusk skylines, light trails, coastlines, and rooftops. Keep the horizon level and the camera move slow.

Food and drink inserts

Coffee pours, steam, ice, and close-up texture. One physical action per clip.

Product inserts

A hand picking up a product, a label turning into light, or a surface detail. Use these between talking-head shots.

Atmospheric cutaways

Rain on a window, fog in trees, curtains moving, or dust in a beam of light. These clips set pace without adding information.

How it works

  1. Write a shot list first

    List the cutaways the edit needs: establishing, texture, transition, and reaction. Generate only the shots on the list.

  2. One action per clip

    A pour, a step, a turn, or a light change. If the prompt contains “and then”, split it into two clips.

  3. Choose a camera move

    Slow dolly, gentle orbit, locked-off, or aerial. A single move makes the clip easy to cut against dialogue.

  4. Match the light

    Name the time of day and direction of light so the b-roll cuts with the main footage instead of fighting it.

B-roll prompt formula

Use this structure as a checklist. Replace the bracketed fields, keep the protection lines, and remove anything that does not apply to your shot.

Create a [duration]-second cinematic b-roll shot of [subject].
Action: [one physical action].
Camera: [one move], [shot size], [lens].
Light: [time of day, direction, quality].
Texture: [rain, dust, steam, fabric, film grain].
No people looking at the camera, no text, no logos, no watermarks.
  • B-roll should not steal focus. Avoid faces looking into the lens unless the shot is a reaction.
  • Name the lens and shot size. “Wide 24mm” and “macro 100mm” produce very different cutaways.
  • Use one texture cue: rain, dust, steam, or grain. More than one turns into visual noise.
  • Generate a little longer than you need so you have handles for the edit.

Model notes

These notes describe where each model tends to fit. They are not a ranking. Model availability changes by provider: WaveSpeed, SeeAPI, FAL, Replicate, and Kie all serve image and video models in this class. The hosted generator shows the available models and credit cost before you run a job.

  • WaveSpeed
  • SeeAPI
  • FAL
  • Replicate
  • Kie
Model fit by workflow, not a benchmark
ModelBest fit
Veo 3.1Detailed environments and coherent lighting for establishing shots.
Sora 2Physically believable water, smoke, and object motion.
Kling 3.0Polished camera moves and cinematic texture.
HailuoStylized motion for social-first edits.
Gemini Omni FlashQuick variations and conversational refinements.

Three prompts you can adapt

Blue-hour city establishing shot

A slow aerial move gives the editor a clean handle at the start and end of the clip.

Create a 5-second cinematic b-roll shot of a city skyline at blue hour with a river and car light trails. Slow forward aerial move over the river toward the skyline. Wide 24mm equivalent, smooth dolly. Low haze, glowing windows, distant traffic. Keep the horizon level and the buildings stable. No text, no logos, no watermarks.

Coffee pour insert

One action, one light source. The steam gives the editor a natural transition point.

Create a 5-second cinematic b-roll shot of coffee pouring into a ceramic cup on a wooden counter. Macro 100mm, 45-degree angle, locked-off with a tiny push-in. Warm morning window light from the left. Steam rises and the liquid settles. Keep the cup, carafe, and counter unchanged. No people looking at the camera, no text, no logos, no watermarks.

Rain on a window

A locked-off texture shot is the easiest clip to cut. It adds mood without competing with dialogue.

Create a 5-second cinematic b-roll shot of rain running down a window at night, with blurred amber and blue city lights beyond. Locked-off close-up, 50mm lens, shallow depth of field. Rain streaks down the glass while the lights pulse softly. Keep the window frame and composition stable. No text, no logos, no watermarks.

Common mistakes

Writing a scene

B-roll is a shot, not a story. If the prompt has a beginning, middle, and end, it belongs in a different workflow.

Too many actions

A pour, a turn, and a smile in five seconds gives the editor nothing clean to cut.

Mismatched lighting

B-roll that does not match the main footage looks inserted. Name the direction and quality of light.

No shot list

Generating without a list produces random clips. Start from the edit and work backward.

FAQ

How long should AI b-roll be?

Five seconds is a practical default. Generate a little longer than the edit needs so you have handles for transitions.

Can I use AI b-roll commercially?

Check the model's commercial terms and your platform's disclosure rules. This guide does not grant rights or provide legal advice.

Do I need a reference image?

Not always. B-roll is a good fit for text-to-video. Use a reference when the shot must match a real location or product.

Which model is best for b-roll?

Use the model that handles your texture best: water and smoke, city aerials, or stylized motion. Compare one prompt across two models.

How many clips should I generate?

Generate three to five variations per shot list item. Pick the cleanest camera move and cut the rest away.

Can I add music?

Yes. Add music in your editor and cut the clips to the beat. Some models can generate audio, but a separate track gives you more control.

Turn the prompt into a clip

Open the hosted generator, paste the prompt, and run one short test. Keep the first generation small; a clean five-second clip is easier to judge than a long one.

Flash Omni is an independent guide and is not affiliated with Google. The hosted generator is a separate product with its own terms and pricing.