The problem with a wall of generated visuals
You wrote a sharp script. The voiceover is clean, the captions are on beat — and yet the video feels flat. Nine times out of ten the culprit is the visuals: every shot is either a static image slowly zooming (the dreaded Ken Burns loop) or a generated clip that's technically impressive but says nothing about your actual point.
B-roll fixes this. B-roll is the supporting footage that plays under your narration — the hands typing, the city street at dusk, the coffee being poured — while your voice carries the idea. Used well, it does three jobs at once: it keeps the frame moving so the viewer doesn't scroll, it makes an abstract point concrete, and it buys you visual variety without a single new sentence of script.
The catch is that most creators either skip B-roll entirely or bury their video in random, meaningless clips. Both feel generic. This post is about the middle path: using real stock footage deliberately, so a short built from a script and a voiceover feels shot, not assembled.
When real footage beats generated video
Generated (text-to-video) clips are genuinely useful — they can show things that don't exist and give you exact control over a scene. But they're slower, cost more per second, and still carry a faint "AI sheen" that some audiences clock instantly. Real stock footage is the opposite: cheap, instant, and unmistakably real.
Reach for stock B-roll when:
- The subject is ordinary and physical. A person walking, a laptop screen, waves, traffic, a kitchen. Real footage of everyday things looks better than any generation and takes a fraction of the time.
- You need pace, not precision. If the shot just has to feel alive under a line of narration — "most people never check their analytics" over someone glancing at a phone — a stock clip is perfect. The viewer isn't studying it; they're listening.
- You're producing at volume. When you're shipping several videos a week, generating every second of footage is a tax on both time and budget. B-roll lets you spend your generation budget only where it earns its keep.
Reach for generated video when the shot is the point — a specific product that doesn't exist yet, a surreal concept, a scene you need to control frame by frame. The best shorts mix both: real footage for the connective tissue, generated shots for the moments that have to be exact.
This "pick the cheapest source that still tells the story" logic is exactly what modern AI video tools are starting to automate. In Vicreon, for example, the pipeline can assemble a montage from real stock footage when a scene only needs to feel right, and reserve pricier generation for the shots that truly need it — so a batch of shorts stays affordable without looking like a stock-photo screensaver.
How to pick B-roll that actually carries the story
The difference between B-roll that works and B-roll that feels like filler comes down to a few habits.
1. Match the shot to the sentence, not the topic. The lazy move is to search your topic ("marketing") and drop in whatever comes up — a generic team high-fiving in an office. The better move is to visualize the specific line you're narrating. If the line is "you're guessing instead of measuring," show someone squinting at a dashboard, not a stock boardroom. One shot, one idea.
2. Prefer motion with a direction. A clip where something moves through the frame — a car passing, a hand reaching, a door opening — reads as intentional. A slow ambient pan reads as a placeholder. Motion gives the cut somewhere to land.
3. Cut before the shot gets boring. In short-form, most B-roll clips should live on screen for one to three seconds. The moment a shot has made its point, cut. A new visual every couple of seconds is a huge part of why some shorts feel "fast" and others feel like a waiting room.
4. Keep a consistent look across the batch. If one clip is warm and cinematic and the next is cold and clinical, the video feels stitched together. Lean toward footage with a similar grade, or apply a light consistent color treatment so a montage of unrelated clips still reads as one piece.
5. Respect the safe zones. Stock footage is usually shot horizontal. When you reframe it to vertical 9:16, keep the important action in the center third — and keep the bottom clear for captions and the platform UI. A gorgeous shot with its subject cropped out of frame is a wasted shot.
The workflow: script first, footage second
The mistake is to go hunting for cool clips and then write around them. Do it the other way:
- Lock the script and voiceover first. The narration is your spine. It decides how many shots you need and how long each one lives.
- Break the script into beats. Every sentence or clause is a potential cut. Note what each beat is literally about — that's your search term.
- Source one clip per beat. Pull real footage for the ordinary beats; flag the two or three beats that genuinely need generated or custom visuals.
- Assemble to the voiceover. Lay the clips under the narration, trim each to its beat, and cut on the natural pauses in the voice.
- Layer captions and a hook frame last. Once the visuals are cutting cleanly, add sound-off captions and make sure the very first shot earns the scroll-stop.
If you're using a tool that automates the mechanical parts — matching footage to script beats, reframing to vertical, timing cuts to the voiceover — this whole loop collapses from an afternoon in an editor to a few minutes of review. That's the real unlock of AI-assisted B-roll: not that a machine picks "creative" shots, but that it handles the tedious sourcing and syncing so you can spend your attention on the two decisions that matter — which beats need custom visuals and does the finished cut hold attention.
Avoiding the "generic stock" trap
The reason "stock footage" has a bad reputation is overuse of the most obvious clips — the laughing salad woman, the abstract data swirl, the anonymous handshake. You avoid it the same way you avoid a clichéd sentence: be specific and be a little unexpected. Search for the concrete noun in your line, not the abstract theme. Favor footage that looks captured rather than staged. And when a clip is doing real narrative work, give it a beat longer to breathe; when it's just texture under a transition, keep it short and move on.
B-roll isn't a downgrade from generated video — it's a different tool for a different job. Use real footage to keep your shorts grounded, fast, and cheap to produce, spend your generation budget on the shots that have to be exact, and let the script decide both. Do that consistently and your AI-made videos stop looking made.