Generating B-Roll with AI (Prompt Library)

TL;DR: B-roll prompting works best organised by the job the clip does in an edit, not by subject. This library gives 40 copy-paste prompts across five jobs, establishing, cutaway, texture, transition-ready and background-for-text, plus the aspect-ratio, audio-default and duration settings that decide which clips actually cut in cleanly.

Prompt Architects writes the prompt, not the video. Nothing on this page renders, edits or publishes a clip; every prompt below is text you send to a video model of your choice, and the result still needs downloading, trimming and dropping into a timeline yourself.

B-roll is the most commercially useful thing an AI video model does right now, more so than a hero shot of a talking product, because nobody scrutinises a three-second cutaway of hands typing the way they scrutinise a face. The bar for "good enough to cut in" is lower, the clip is shorter, and a slightly odd frame in the corner rarely survives long enough on screen to matter. That combination is why B-roll prompting is worth a library of its own rather than a handful of examples bolted onto a general video-prompting guide.

It is also the part of an edit where an editor already expects to reach outside the main footage. A shot of a product founder talking to camera has always been supplemented with stock footage, a licensed clip, a screen recording, something shot separately and cut in around the interview. AI video slots into a gap that was never going to be the original footage anyway, rather than trying to replace the one shot everyone is actually watching. That is a materially easier bar to clear than "generate the whole video," and it is why this is worth treating as its own skill rather than a footnote in a general prompting guide.

What Job Is Each Type of B-Roll Actually Doing?

Five jobs cover almost everything an editor reaches for B-roll to do, and each wants a different kind of clip. Naming the job before you write the prompt changes what you put in it: a clip built to establish a place answers "where are we," a clip built as a cutaway answers "what else is happening," and neither question is well served by a prompt written for the other.

• Establishing sets a location or scale before the main subject appears, usually wide, usually static or slow, and it earns its place by giving the viewer somewhere to stand before anything else happens.
• Cutaway shows a related detail or action to cover a jump cut, usually medium or close, with a clear single action, since a cutaway with two competing actions reads as indecisive rather than as coverage.
• Texture is mood and material, an abstract or macro detail meant to be felt rather than read, often held longer or looped, and it works precisely because there is nothing in frame demanding to be identified.
• Transition-ready clips are built to end on a frame that bridges cleanly into the next shot, a whip pan, a closing door, a filled frame, so the cut lands on motion rather than on a static hold.
• Background-for-text sits behind a caption or lower-third, deliberately low on competing detail and slow enough that the words stay readable, since anything busier competes with the one thing the viewer actually needs to read.

Getting the job right matters more than getting the subject right. A gorgeous wide shot is a bad cutaway if it pulls focus from the caption sitting over it, and a busy macro texture shot is a bad opener if it never establishes where the viewer is. The reverse mistake is just as common: a plain establishing shot dropped in as a cutaway reads as filler rather than as coverage, because it was never built to hold a viewer's attention for its own sake. If you have not built a full structured prompt before, post 205 walks the seven-part anatomy that every prompt below follows in miniature: subject, action, setting, camera, light, look, and where relevant, sound.

Establishing Shots: 8 Prompts to Open a Scene

Wide, usually static or slow-moving, no character mid-action required. These want a beat of stillness so the eye can read the space before the edit moves on, which is also why a slow push-in or a gentle rise reads better here than a whip pan would: the motion should reveal scale, not compete with it.

1. Aerial drone shot descending slowly over a coastal city skyline at golden hour, warm
light on glass towers, faint haze over the water, smooth continuous descent with no
camera shake.

2. Wide static shot of an open-plan office floor at 8am, empty desks, morning light
cutting long shadows through half-open blinds, dust motes visible in the light beams.

3. Wide shot of a highway overpass at dusk, red and white traffic-light trails streaking
below on a slow shutter, city skyline in silhouette behind it, a gentle push-in.

4. Wide shot of a farmers market stall photographed from across the street, colourful
stacked produce in the foreground, soft overcast morning light, background pedestrians
out of focus.

5. Aerial shot rising slowly over a manufacturing plant rooftop at sunrise, steam lifting
from vents, orderly rows of solar panels catching the first light.

6. Wide static shot of a hospital entrance at dawn, an empty ambulance bay, soft blue
pre-sunrise light, automatic doors sliding open and closed on their own cycle.

7. Wide shot of a university lecture hall filling with students before class, high angle
from the back row, natural window light, no dialogue audible.

8. Wide shot of a shipping port at sunset, container cranes in silhouette against an
orange sky, gentle reflections moving on the harbour water.

Cutaway Shots: 8 Prompts for a Believable Aside

Medium to close, one clear action, framed to survive being on screen for two or three seconds and no longer. A cutaway does not need a beginning, a middle and an end the way a hero shot does; it needs one continuous action a viewer can read at a glance, which is why every prompt below names a single verb rather than a small sequence of them. Post 211's camera-movement reference is worth pairing with these if you want to vary the framing beyond what is written below.

1. Close-up of hands typing on a laptop keyboard, shallow depth of field, soft daylight
from an office window, the screen itself out of focus behind the hands.

2. Medium shot of a barista steaming milk at an espresso machine, steam rising visibly,
warm café lighting, hands and the steel pitcher in sharp focus.

3. Close-up of a hand signing a printed document with a pen, paper texture visible under
raking desk light, the signature completing in one continuous motion.

4. Medium shot of two coworkers shaking hands across a conference table, background
softly blurred, natural daylight through office windows behind them.

5. Close-up of a smartphone screen showing a notification banner appear, a hand holding
the phone steady, background out of focus.

6. Medium shot of a warehouse worker scanning a barcode on a box with a handheld
scanner, even industrial overhead lighting, a quick natural motion.

7. Close-up of a chef plating a dish, garnish placed with tweezers, warm kitchen
lighting, shallow depth of field isolating the plate.

8. Medium shot of a delivery driver checking a clipboard beside a parked van, flat
overcast daylight, framed from the waist up.

Texture Shots: 8 Prompts for Mood and Detail

Abstract or macro, meant to be felt rather than read, and often the clip that gets held on screen the longest or looped behind other footage. Because nothing in frame needs identifying, texture shots tolerate the small drift and softness that would sink a cutaway, which makes them a forgiving place to start if you are new to prompting for video specifically.

1. Extreme close-up of rain droplets sliding down a window pane, city lights reduced to
soft bokeh in the background, slow-motion, static camera.

2. Macro shot of steam curling off a cup of coffee on a wooden table, soft warm side
light, shallow focus on the rising steam.

3. Close-up of the woven texture on a wool sweater, natural side light raking across the
fibres, completely static camera.

4. Macro shot of ink diffusing into still water, slow-motion, dark background, a single
coloured plume unfurling outward.

5. Close-up of tree leaves moving in a light wind, dappled sunlight flickering through
the canopy, subtle handheld movement.

6. Macro shot of a match striking and igniting, slow-motion, dark background, the spark
trail clearly visible.

7. Close-up of dry sand sifting through fingers, backlit at golden hour, shallow depth of
field on the falling grains.

8. Macro shot of condensation beading on a cold glass surface, static camera, soft even
studio lighting.

Transition-Ready Shots: 8 Prompts Built to Bridge a Cut

Framed to end on something that swallows the frame or fills it completely, so the next clip can start from that same visual anchor. None of these ask the video model to perform an actual cut between two scenes; that part still happens in your editor. What the prompt controls is the last frame the clip hands off, a door, a hand, a wall of motion blur, which is what makes the edit read as intentional rather than as two unrelated clips placed next to each other. Post 322 covers cutting technique between two clips in full if you want the editor-side half of this job.

1. Quick whip pan across a city street at eye level, motion blur through the pan,
natural daylight, ending on a static hold of a storefront window.

2. Medium shot of a door opening from a dark hallway into a bright room, the light
expanding to fill the frame by the end of the shot.

3. A person walking directly toward the camera until their body fills the frame in
darkness, natural even walking pace, no camera movement.

4. Overhead shot of a hand placing a folder onto a desk, the folder's surface briefly
covering the lens as it settles into frame.

5. Fast dolly-in on a spinning ceiling fan, motion blur increasing as the camera
approaches, natural room lighting.

6. A car door closing directly toward the camera, the door's surface filling the entire
frame in the final moment, natural daylight.

7. Time-lapse of clouds moving quickly across a blue sky above a rooftop, useful as a
short scene-bridging beat between two locations.

8. Close-up of a curtain being drawn across a window, the fabric filling the frame by
the end of the motion, soft interior light.

Background-for-Text Shots: 8 Prompts That Won't Fight Your Captions

Deliberately low on competing detail, slow, and quiet enough that on-screen text stays readable over the top. The test for this category is simple: read the caption you plan to place over the clip out loud while picturing the motion behind it, and if the motion is the thing your attention keeps snagging on, the shot is too busy for the job. Post 294 covers looping this kind of clip properly if you need it to repeat without a visible jump.

1. Slow, gentle drift over out-of-focus city lights at night, soft bokeh circles shifting
position, muted blue and amber tones, no sharp edges anywhere in frame.

2. Soft, slow-moving gradient of clouds against a pastel sky, static camera, a gentle
colour shift, no foreground subject to compete with overlaid text.

3. Blurred office background with soft bokeh from desk lamps, very slow camera drift,
low contrast throughout, framed for a lower-third caption.

4. Abstract macro shot of paint slowly mixing into water, soft pastel colours,
slow-motion, no sharp detail, calm continuous movement suitable for a looping headline.

5. Wide, softly blurred shot of a beach at low tide, muted colours, gentle repetitive
wave motion, no people in frame.

6. Slow pan across a softly blurred bookshelf, warm library lighting, shallow depth of
field, gentle continuous motion with no sharp text-competing detail.

7. Abstract macro shot of light refracting through a prism, slow colour drift, dark
background, calm motion suitable for a title card.

8. Soft, out-of-focus shot of falling snow against a dark background, gentle continuous
motion, contrast kept low throughout.

Which Settings Actually Matter for B-Roll?

Three fields decide whether a clip is usable for the job you wrote it for, and none of them are about the words in the prompt.

Sources, accessed September 3, 2026: kling.ai/document-api/api/video/3-0-omni/text-to-video · ai.google.dev/gemini-api/docs/veo · docs.dev.runwayml.com/api.md.

Aspect ratio decides whether the same prompt is reusable across a square social feed, a vertical Reel and a 16:9 landing-page hero, or whether you need three separate generations. Kling 3.0 Omni is the only one of these three with a native 1:1, "Veo 3.1 lets you create landscape (16:9, the default setting) or portrait (9:16) videos" (ai.google.dev/gemini-api/docs/veo, read September 3, 2026), and Runway's gen4.5 branch caps out at 1280:720 or 720:1280, both 720p-equivalent, with no separate resolution field at all. If you need genuine 4K from Runway, that comes from a separate POST /v1/video_upscale endpoint billed per output frame, not from the generation call itself.

Audio default matters most for the background-for-text and texture jobs, where a clip usually sits under a voiceover or a music bed you have already chosen. Kling's settings.audio field defaults to off, so the clip arrives silent unless you deliberately set it to native. Veo bakes audio into every current generation with no field to turn it off, and "Video extension is also limited to 720p videos" (ai.google.dev/gemini-api/docs/veo, read September 3, 2026) on top of that, so a Veo clip you plan to extend later loses both its audio-free option and its top resolution at once. Runway's gen4.5 parameter list has no audio field either, an absence rather than a documented toggle.

Duration matters most for establishing and cutaway shots, which rarely need to run longer than the cut itself. Post 234 has the complete duration table across eleven surfaces, including which models couple a longer duration to a lower resolution, a trade you want to know about before you generate ten clips at the wrong length.

How Do You Actually Turn These Into Something You Can Cut Into a Timeline?

The same way as any other AI video output: paste the prompt into whichever model you have access to, set the aspect ratio and duration for the job at hand, generate, download, and bring the file into your editor. Nothing here shortcuts that pipeline, and no vendor checked for this page offers a way to skip it either.

A few honest constraints are worth knowing before you generate a batch. First, consistency across clips is not guaranteed: if you need the same background element or product to reappear in two separate B-roll shots, post 227 ranks what actually helps, and none of the options are a single setting. Second, a clip that repeats without a visible jump is the exception, not the default, on every model except Luma; treat a clean repeat as something you build with a loop-capable model or trim for in the edit, not something any prompt guarantees on its own. Third, shorter is usually safer for the establishing and cutaway jobs specifically: a 3 to 6 second clip has less room for the motion to drift into something unusable than a 10 or 15 second one, and it is cheaper to regenerate if it does.

Put together, these three constraints point at the same underlying discipline: generate short, generate for one job at a time, and treat every clip as a candidate for the timeline rather than a guaranteed placement. A batch of six establishing shots at 4 seconds each, reviewed and culled down to the two that actually work, is a more reliable workflow than one 15-second generation you are hoping to make do.

None of the 40 prompts above are precious. Swap the subject, keep the job-specific framing and pacing intact, and the category still does its work in the edit. The pacing and timing post covers the words that control how a clip feels once you're past picking the right shot, which matters most for the texture and transition-ready jobs above, where the motion itself is most of what the clip is doing.