Veo 3 prompt guide: shots, camera, audio and dialogue

Veo 3 is Google's video generation model family, and its standout feature is native audio: it can generate sound effects, ambience, music and spoken dialogue in sync with the picture from a single prompt. That makes a Veo 3 prompt closer to a short film brief than an image prompt. You are directing the camera, the actors and the sound at once. This guide walks through each part of a strong prompt, based on Google's official prompting guidance for Veo, with examples you can adapt.

How Veo 3 reads a prompt

Google's prompting guide for Veo 3.1 sums up a reliable structure as a formula:

[Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance]

Cinematography is the camera work and framing; subject is the main character or focal point; action is what happens; context is the environment; style and ambiance cover the look, mood and lighting. Audio and dialogue are added as separate, clearly marked sentences. Here is Google's own example of the visual part:

Medium shot, a tired corporate worker, rubbing his temples in exhaustion,
in front of a bulky 1980s computer in a cluttered office late at night.
The scene is lit by the harsh fluorescent overhead lights and the green
glow of the monochrome monitor. Retro aesthetic, shot as if on 1980s color
film, slightly grainy.

Keep the clip length in mind. Google lists clips of 4, 6 or 8 seconds for Veo 3.1, in 16:9 or 9:16. Eight seconds is enough for one clear action and perhaps a line of dialogue, not a whole story. If your prompt describes five events, Veo has to compress or skip some of them.

Shot and camera movement

Start with how the scene is filmed. Google groups camera language into three families, and using the standard terms gives you much more control than "make it cinematic":

• Composition: wide shot, close-up, extreme close-up, low angle, two-shot.
• Movement: dolly shot, tracking shot, crane shot, aerial view, slow pan, POV shot.
• Lens and focus: shallow depth of field, wide-angle lens, soft focus, macro lens, deep focus.

Crane shot starting low on a lone hiker and ascending high above,
revealing they are standing on the edge of a colossal, mist-filled canyon
at sunrise, epic fantasy style, awe-inspiring, soft morning light.

Describe the movement as a path with a start and an end ("starting low… ascending high above, revealing…"). One deliberate camera move per clip usually looks better than combining a pan, a zoom and an orbit.

Subject, action and setting

Be specific about who or what is on screen. "A man" gives Veo little to hold on to; "an elderly fisherman in a yellow raincoat with a grey beard" gives it a character. Specific details also help when you generate several clips and want the character to look consistent.

Write the action with concrete verbs and a clear sequence. "She turns, notices the letter on the table, and slowly picks it up" is easier to animate than "she reacts emotionally". For the setting (Google calls it context), add the time of day, weather and a few background elements that make the place feel real.

Close-up with very shallow depth of field, a young woman's face, looking
out a bus window at the passing city lights with her reflection faintly
visible on the glass, inside a bus at night during a rainstorm, melancholic
mood with cool blue tones, moody, cinematic.

If there are several people, identify each one by a distinctive detail ("the man in the red hat", "the woman with the green scarf"). Google's Vertex AI guide recommends this, especially when you later assign actions or lines to them.

Style, lighting and mood

Style can be general ("cinematic", "documentary", "stop-motion animation") or very specific ("shot as if on 1980s color film, slightly grainy"). Lighting does a lot of the emotional work: harsh fluorescent light reads as tired and clinical, golden hour as warm and nostalgic, neon as urban and electric. Add a mood word and a color direction ("cool blue tones", "warm amber palette") to tie the look together.

Avoid stacking contradictory styles. "Photorealistic anime watercolor" forces the model to guess which one you meant. Pick one visual language and support it with lighting and color.

Audio: sound effects, ambience and music

Audio is what makes Veo 3 different, and Google's guidance gives it clear syntax. Describe sound in separate sentences rather than mixing it into the visual description:

• Sound effects: SFX: thunder cracks in the distance
• Ambient noise: Ambient noise: the quiet hum of a starship bridge
• Music: describe it as a cue, for example SFX: a swelling, gentle orchestral score begins to play, or a plain sentence such as "Soft acoustic guitar music plays in the background."

Wide shot of a small harbor at dawn, fishing boats rocking gently at their
moorings, mist over the water, soft pink light.
SFX: wooden hulls creaking, a rope knocking against a mast.
Ambient noise: distant gulls and lapping waves.
Calm, slow piano music plays quietly in the background.

Be concrete about sounds the way you are about visuals: "footsteps on wet gravel" is better than "walking sounds". If a sound should happen at a specific moment, tie it to the action ("as the door slams, the lights flicker").

Dialogue in quotes

For speech, Google's rule is simple: put the exact words in quotation marks and attribute them to a character.

Medium shot in a dim detective's office, rain against the window. A weary
detective behind his desk looks up at a woman in a green coat and says in a
low, tired voice, "Of all the offices in this town, you had to walk into
mine." The woman smiles slightly.

• Keep lines short. In an 8-second clip, one or two short sentences fit naturally; long speeches get rushed or cut.
• Describe the delivery: "whispers", "shouts over the wind", "says cheerfully".
• Say who speaks when there are several characters, using their distinguishing details.
• Unwanted captions: a common community practice is to add "(no subtitles)" after the dialogue, or to include "subtitles, captions" in the negative prompt, since some clips come out with burned-in text.

Negative prompts: what to leave out

Google's advice on exclusions comes in two forms. In the main prompt, describe what you want instead of what you do not want: its example is "a desolate landscape with no buildings or roads" rather than "no man-made structures". Where a separate negative prompt field is available, as in some API and studio interfaces, Google's Vertex AI guide says to simply list the unwanted elements without instructive words like "no" or "don't".

Negative prompt: subtitles, text overlays, watermark, urban background,
man-made structures, distorted hands

Keep negative prompts short and targeted at problems you have actually seen. A long generic list adds noise without solving anything specific.

Image-to-video, reference images and timed shots

Veo can also start from images, which gives you far more control over the look:

• Image-to-video: animate a still. Describe only the motion, camera and sound; do not re-describe a different scene. Make sure the action matches what is visible in the image.
• Ingredients to video: provide reference images of a character, object, setting or style to keep them consistent across several shots. Start prompts with "Using the provided images for…".
• First and last frame: give a start image and an end image, and describe the transition. Google's example: "The camera performs a smooth 180-degree arc shot, starting with the front-facing view of the singer and circling around her to seamlessly end on the POV shot from behind her on stage."

Google's guide also shows timestamp prompting, which directs several shots inside one generation. Here is a condensed version of its example:

[00:00-00:02] Medium shot from behind a young explorer pushing aside a large
jungle vine to reveal a hidden path.
[00:02-00:04] Reverse shot of her face, filled with awe as she sees ancient,
moss-covered ruins. SFX: rustling leaves, distant bird calls.
[00:04-00:06] Tracking shot as she steps into the clearing and touches the
carved stone wall.
[00:06-00:08] Wide, high-angle crane shot revealing the vast temple complex.
SFX: a gentle orchestral score begins.

Feature availability differs between the Gemini app, Flow and the API, so check which options your interface offers. Google marks all Veo videos with a SynthID watermark to identify them as AI-generated.

Build the prompt faster

Our Veo 3 prompt generator runs in your browser and assembles these pieces from a form: shot, camera move, scene, action, style, lighting, mood, an audio field written as an SFX cue, dialogue placed in quotes with "(no subtitles)", and a negative prompt line. For Veo, choose 8 seconds and 16:9 or 9:16, since those match Google's listed options; the other durations in the menu are meant for other models. To prepare a strong start frame, write it with the Nano Banana prompt generator, and to compare Veo with other models, try the video prompt generator, which also outputs prompts for Sora 2, Kling, Runway and more.