Veo 3 rewards specific, structured prompts far more than older video models did. Because it generates picture and native audio in a single pass — ambience, sound effects, even spoken dialogue — a good Veo 3 prompt reads less like a caption and more like a line from a shooting script: who is on screen, what they do, where they are, how the camera moves, and what you should hear.
This guide gives you the full formula. First the anatomy of a strong prompt, then nine copy-ready examples across the genres people actually search for — cinematic shots, product spots, ASMR, talking-head vlogs, anime style, aerials — each with a note on why it works. After that: negative prompts, image-to-video and keyframe prompting, and the common mistakes that make Veo ignore half of what you wrote.
Every prompt below runs unchanged on Veo AI, where Veo 3.1, Veo 3.1 Fast, and Veo 4 are available in the browser with free starter credits — so you can paste an example into the studio and test it before writing your own.
Anatomy of a strong Veo 3 prompt
Think of a Veo 3 prompt as eight slots, filled in roughly this order: subject, action, setting, camera move, shot type, lighting, style, and audio. You rarely need all eight — four or five well-chosen ones beat eight vague ones — but the order matters, because Veo weights what comes first. Lead with the subject and action; save stylistic flourishes for the end.
The audio slot is what sets Veo apart. Describe the soundscape the way you describe the picture: room tone, footsteps, a specific song genre, wind. For speech, put the exact line in quotation marks and attribute it — Veo lip-syncs quoted dialogue. A prompt with no audio direction gets generic sound, which is the single most common reason clips feel flat.
- Subject — who or what, with concrete detail: ‘a silver-haired woman in a yellow raincoat’, not ‘a person’.
- Action — one clear verb phrase per clip; Veo handles one beat well, three beats badly.
- Setting — place, time of day, weather. This drives lighting more than the lighting slot does.
- Camera move — dolly in, slow pan, handheld follow, static tripod. Pick exactly one.
- Shot type — close-up, medium, wide, aerial, over-the-shoulder.
- Lighting — golden hour, neon, overcast softbox, single practical lamp.
- Style — ‘shot on 35mm film’, ‘anime style’, ‘hyperrealistic commercial’.
- Audio — ambience, effects, music genre, and any dialogue in quotes.
Veo 3 prompt examples you can copy
Nine templates across the genres that dominate Veo 3 searches. Paste one into the studio, swap the nouns, and iterate from a result instead of a blank box. Each stays inside the formula above — notice how short the strong ones are.
Cinematic drama
‘A weathered fisherman hauls a rope onto a wooden dock at dawn, thick fog over the water. Slow dolly-in to a medium close-up. Cold blue light, shot on 35mm film. Audio: creaking rope, distant gulls, low ocean swell.’
Why it works: one action, one camera move, and an audio bed that matches the mood — Veo fills the fog and film grain convincingly when the lighting and lens are named.
Product commercial
‘Macro shot of a matte-black wireless earbud rotating on a white marble pedestal, soft studio lighting with a slow orbiting camera. Water droplets bead on the surface. Style: premium tech commercial. Audio: minimal electronic music, a subtle whoosh as the earbud turns.’
Why it works: product prompts live or die on surface detail and controlled light; ‘macro’, ‘matte’, and ‘marble’ give Veo textures it renders exceptionally well.
ASMR
‘Extreme close-up of gloved hands slicing a glossy honeycomb block on a slate board, slow and deliberate. Static camera, shallow depth of field, warm soft light. Audio: rich sticky slicing sounds, faint crackle, no music, no talking.’
Why it works: the veo 3 asmr prompt genre is pure audio — explicitly banning music and speech forces Veo to foreground the tactile sound.
Vlog-style talking head
‘Selfie-style handheld shot of a cheerful hiker on a mountain trail, slightly out of breath, looking into the camera and saying: “Okay, we just hit the summit and this view is unreal.” Golden hour backlight, wind in the microphone, birds in the distance.’
Why it works: quoted dialogue plus ‘selfie-style handheld’ reliably produces the viral first-person vlog look, with lip-sync on the spoken line.
Street interview
‘A documentary street interview on a busy city sidewalk at dusk. A reporter holds a microphone to an elderly man in a flat cap who chuckles and says: “In my day, we just called it dancing.” Medium two-shot, handheld, neon signs bokeh behind them. Audio: traffic hum, crowd chatter.’
Why it works: the street-interview format is one of the most-searched Veo 3 prompt templates; naming both characters and giving only one of them a line keeps the lip-sync clean.
Animal character
‘A capybara in a tiny knitted scarf sits in an outdoor hot spring at night, steam rising, snow falling gently. Slow push-in to a close-up as it closes its eyes contentedly. Cozy cinematic lighting. Audio: bubbling water, soft snowfall ambience, a quiet satisfied sigh.’
Why it works: animal clips go viral on charm, not realism — one anthropomorphic detail (the scarf) and an emotional beat (the sigh) do the work.
Anime style
‘Anime style: a schoolgirl runs across a pedestrian bridge at sunset, cherry blossom petals streaming past, her ribbon trailing. Dynamic tracking shot alongside her, dramatic orange-and-pink sky, 2D cel-shaded look with painterly clouds. Audio: emotional piano, wind, her hurried footsteps.’
Why it works: leading with ‘anime style’ and reinforcing it with ‘cel-shaded’ stops Veo from drifting back toward photorealism mid-clip.
Aerial landscape
‘Aerial drone shot flying low over a turquoise glacial river braiding through black volcanic sand, Iceland. The camera rises and tilts up to reveal distant snow-capped peaks. Overcast diffused light, hyperrealistic. Audio: deep ambient drone music, rushing water far below.’
Why it works: a single continuous camera gesture — fly low, rise, tilt — reads as one planned move; aerials fail when the prompt asks for several disconnected angles.
Music-driven montage
‘A breakdancer spins on wet asphalt under a highway overpass at night, puddles reflecting sodium-orange streetlights. Low-angle shot, the camera circling him in one smooth move. Audio: heavy boom-bap hip-hop beat synced to his movements, cloth snapping, echoing space.’
Why it works: telling Veo the music genre and asking for movement ‘synced’ to it produces rhythm you cannot get by adding a track afterwards.
Prompting audio and dialogue
Veo 3 generates sound natively, so treat audio as a first-class part of every prompt rather than an afterthought. Three layers are worth directing separately: ambience (room tone, weather, traffic), effects tied to the action (footsteps, pouring, impacts), and music (genre and mood, not a song title). If you want silence or no soundtrack, say so — ‘no music’ is a legitimate and useful instruction.
For speech, the rule is simple: put the exact words in quotation marks and say who speaks them. Unquoted descriptions like ‘she talks about her day’ produce mumbled, non-lip-synced murmuring — which is the answer to the perennial ‘why is there no talking in my Veo 3 prompt’ question. Keep lines short; a 4–8 second clip comfortably fits one sentence of dialogue, not a monologue. On Veo AI this works across Veo 4 and the Veo 3.1 tiers, with the exception of Veo 3.1 Lite, which renders video only.
Negative prompts: what to exclude
A negative prompt lists what you do not want, and Veo supports it as a separate field — use it instead of writing ‘no subtitles’ into the main prompt, where the word ‘subtitles’ can ironically summon them. State the unwanted thing as a plain noun: ‘subtitles, captions, text overlays, watermark’ is a near-universal starter set.
Beyond text artifacts, negatives are most useful for style policing. Generating anime? Negative ‘photorealistic, live action’. Shooting a clean product spot? Negative ‘hands, people, cluttered background’. Keep the list short and concrete — negative prompts steer composition, they do not fix a vague main prompt.
Image-to-video and keyframe prompting
When you start from an image, the picture carries the subject, setting, lighting, and style — so your text prompt should stop re-describing them and spend its budget on motion and sound instead. ‘The camera slowly pushes in as steam rises from the cup; soft café chatter’ is a complete image-to-video prompt. Re-describing the scene in words that slightly contradict the image is the fastest way to get warping.
Veo on this platform also accepts first-and-last-frame keyframes and reference images. With keyframes, prompt the journey between the frames — the transformation, the camera path — rather than either endpoint. Reference images solve character consistency across multiple clips: lock the face once, then vary the action per generation. Combined with clip extension, this is how prompt-driven multi-shot sequences stay coherent.
Common Veo 3 prompt mistakes
Most failed generations trace back to a handful of patterns. If Veo ‘never gets your prompt right’, it is almost always one of these rather than the model ignoring you:
- Overstuffing — three actions, two locations, and a plot twist in one 8-second clip. One beat per generation; chain clips for stories.
- Contradictory camera directions — ‘static shot, the camera sweeps around’. Pick one move and commit.
- No audio direction — you get stock ambience and no speech. Direct the sound every time.
- Dialogue without quotes — paraphrased speech produces murmur instead of lip-synced lines.
- Style buried at the front — leading with ‘in the style of…’ before the subject weakens both.
- Fighting the image in image-to-video — restating (or contradicting) what the frame already shows.
- Vague nouns — ‘a nice car in a cool place’ gives Veo nothing to render; specificity is the whole game.
Test your prompts on Veo AI
Prompt skill compounds through reps, and reps are cheaper here: sign up for free starter credits, paste any example above into the studio, and run it on Veo 3.1 — or on Veo 4, Veo 3.1 Fast, and the Lite tier when you want faster, cheaper iterations while drafting. You can also A/B the exact same prompt against Sora 2, Kling, or Hailuo to see which engine suits your style, then download the winner watermark-free.