A Veo 3 prompt guide built only from Google's documentation: the five-part formula, dialogue and audio syntax, camera terms, negative prompts and templates.
Method. Google's Veo prompt guide, best practices, model, pricing and Gemini API pages read November 18, 2026; Veo 3.1 confirmed as the current version, and the templates are our untested illustrations. Vinc makes an AI film studio.
Veo 3 prompt guide: the short answer
A good Veo prompt names the shot, the subject, the action, the place and the look, in that order. Audio goes in separate sentences, and anything you don't want goes in the negative prompt field as plain nouns. Google's published formula is [Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance], from its Veo 3.1 prompting guide (16 October 2025).
Everything below comes from Google's own documentation. For how Veo's rules compare with Kling, Runway, LTX and others, see How to Prompt AI Video Models.
Which Veo version is current in November 2026?
Veo 3.1 is the current Veo model. On 18 November 2026, Google DeepMind's Veo page leads with it, and the Agent Platform model page lists three Veo 3.1 models and nothing later. We found no official Google page for a Veo 4. Google's prompting rules are the same for 3 and 3.1.
All three output 16:9 or 9:16; for delivery specs, see Aspect Ratios Explained. Google's Gemini API video docs say the Veo 3.1 preview models in the Gemini API were shut down on 22 October 2026. And the model page lists audio generation but not audio input, so you can't hand Veo a song or voice track; add licensed music in the edit (see Music Licensing for Film, Explained).
Cinematography: "the camera work and shot composition"
Subject: "the main character or focal point"
Action: "what the subject is doing"
Context: "the environment and background elements"
Style & ambiance: "the overall aesthetic, mood, and lighting"
Google's five-part Veo 3.1 formula (cinematography, subject, action, context, style and ambiance), in the order its worked example follows (Google Cloud blog prompting guide).
Its worked example follows the formula in order:
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.
Google's video generation prompt guide adds lens, temporal elements and audio, and says you don't need every element every time. The DeepMind prompt guide suggests stating the style (realistic, animated, stop-motion) first.
Google's best practices page says to "dedicate each prompt to a single, focused moment", warning that chaining events "often leads to muddled or incomplete videos". With 8 seconds maximum, one action per prompt is the safe default.
How do you write dialogue and audio in a Veo 3 prompt?
Google says to "clearly specify if you want audio" and to describe it in separate sentences, covering sound effects, ambient noise and dialogue. The Cloud blog guide uses labels such as SFX: thunder cracks in the distance and Ambient noise: the quiet hum of a starship bridge; the DeepMind guide uses a separate Audio: line.
Dialogue has two published syntaxes, and they conflict. The October 2025 Cloud blog guide says to "use quotation marks for specific speech". Google's current best practices page says to put a colon after the speaker and skip quotation marks "To prevent the model from rendering text in the video":
A woman says: My name is Clara.
The current prompt guide uses the colon form throughout, and writes delivery as plain description, such as "a voiceover with a polished British accent speaks in a serious, urgent tone". If quoted lines turn into burned-in captions, switch to the colon form. Google's DeepMind page also says natural spoken audio, especially in short speech segments, is still under development.
Which camera terms does Veo understand?
Google's prompt guide lists these terms by name, with the caveat that some advanced angles and lenses are "not officially supported" and that "The results and reliability may vary".
Category
Terms in Google's guide
Angles and shot sizes
Eye-level, low-angle, high-angle, bird's-eye or top-down, worm's-eye, Dutch or canted, close-up, extreme close-up, medium shot, full or long shot, wide or establishing shot, over-the-shoulder, point-of-view
Movements
Static or fixed, pan (left/right), tilt (up/down), dolly (in/out), truck (left/right), pedestal (up/down), zoom (in/out), crane, aerial or drone, handheld or shaky cam, whip pan, arc
Lens and optics
Wide-angle, telephoto, shallow depth of field (bokeh), deep depth of field, lens flare, rack focus, fisheye, vertigo effect (dolly zoom)
The Veo API has a separate negativePrompt parameter, "a string value that describes content that you want to prevent the model from generating" (image-to-video reference).
List the unwanted things; don't give instructions. Google marks "No walls" and "don't show walls" as not recommended, and "wall, frame" as recommended. In the main prompt, describe positively, for example "a desolate landscape with no buildings or roads" rather than "no man-made structures".
How do you prompt Veo image-to-video, first/last frame and reference images?
Veo 3.1 takes three kinds of image input, and Google advises differently for each.
Image-to-video (first frame). The image already supplies subject, scene and style, so "prompt for motion only"; "Redundant prompts confuse the model and lead to poor results." Call the person "the subject" or "they", and describe motion such as "Slow dolly in on the subject". Start from a sharp image of at least 720p in 16:9 or 9:16.
First and last frame. Veo generates "a natural video transition between a provided start image and end image, complete with audio", so the prompt describes only the move between the two.
Reference images ("ingredients to video"). You supply a character, object, scene or style "to maintain a consistent aesthetic across multiple shots", and name each by role in the prompt. Reference-image-to-video only supports 8-second clips.
Can you make multi-shot videos with timestamp prompting?
Yes. Timestamp prompting is the Cloud blog guide's way "to direct a complete, multi-shot sequence with precise cinematic pacing, all within a single generation". Google's example splits 8 seconds into four 2-second shots, starting:
[00:00-00:02] Medium shot from behind a young female explorer with a leather satchel and messy brown hair in a ponytail, as she pushes aside a large jungle vine to reveal a hidden path.
Google's timestamp-prompting example: one 8-second generation split into four 2-second shots, each written as its own timestamped line.
Later lines add SFX: and Emotion: labels. This partly contradicts the "single, focused moment" advice, so use it when the cuts are the point, as in a montage, and keep each segment to one action.
Veo 3 prompt templates
These are our illustrations of the documented rules above, untested against the model. Fill the bracketed slots.
1. Single shot with ambient sound:
[Shot size and angle], [subject with 2–3 visible details], [one action], in [place, time of day, weather]. [Lighting], [mood], [style or film stock].
Ambient noise: [background sound]. SFX: [one distinct sound tied to the action].
Negative prompt: [unwanted nouns, comma-separated]
2. Dialogue line (colon form):
Medium close-up, eye level, of [character description] in [place]. [He/She] looks at [other character or camera] and says: [line of dialogue].
Ambient noise: [room tone or location sound].
Negative prompt: subtitles, text, music
3. Image-to-video, motion only:
Slow dolly in on the subject. The subject [small action]. [Environmental motion, e.g. dust drifts through the light].
Audio: [one or two sounds].
4. First and last frame transition:
The camera performs a [move, e.g. slow arc / crane up], starting on [what the first frame shows] and ending on [what the last frame shows]. [Any action during the move].
5. Consistent character across shots:
[Character name], a [age, build, hair, eye colour, signature clothing], speaking in a [voice style]. [New shot size and camera move]. [Character name] [new action] in [new setting].
6. Four-shot montage in one generation:
[00:00-00:02] [Shot 1 size and move], [subject], [action].
[00:02-00:04] [Shot 2], [action]. SFX: [sound].
[00:04-00:06] [Shot 3], [action]. Emotion: [feeling].
[00:06-00:08] [Shot 4, often a wide or crane reveal], [action]. [Music cue].
How do you keep a character consistent across Veo shots?
Google's best practices page gives two methods:
Fixed description. Give the character a name, voice style and fixed features, and "copy and paste the entire, unchanged character description into your prompt for every new scene or action". Change only the action or setting.
Same seed. A seed "without changing other parameters guides the model to produce the same videos" (range 0 to 4,294,967,295), keeping visual, stylistic and voice output consistent.
Reference images do the same job visually; for wardrobe, see Costume and Wardrobe Continuity. Vinc's studio says you "lock a face, wardrobe or set once; every shot keeps it."
What does a Veo 3 clip cost?
Google's Agent Platform pricing lists Veo by resolution, with and without audio, at the same price for text and image prompts.
Veo 3.1 list price per second, video + audio, 1080p
Veo 3.1$0.40
Veo 3.1 Fast$0.12
Veo 3.1 Lite$0.08
Google Cloud Agent Platform pricing, checked 18 Nov 2026 (Google lists the unit as 'count'; Runway's API lists Veo 3.1 at 40 credits, $0.40, per second). Video-only output is cheaper: $0.20, $0.10 and $0.05. Veo 3.1 Lite is in Preview.
So one 8-second 1080p clip with audio costs $3.20 on Veo 3.1, $0.96 on Fast and $0.64 on Lite per attempt (our arithmetic). 4K costs $0.60 per second with audio on standard and $0.30 on Fast. A shot that takes four tries costs four times as much, so prompt discipline is also a budget tool.
Vinc (us) offers Veo 3.1 Fast at 18 credits a second with audio and 12 without, $0.18 and $0.12 at the Starter plan's $0.01 a credit by our arithmetic (Vinc pricing). For other models, see AI Video Models Compared; for the full cost of a shot, What an AI Shot Really Costs.
How we compiled this
We read Google's Veo documentation, pricing and DeepMind pages on 18 November 2026, plus the Google Cloud blog's 16 October 2025 prompting guide, and used no third-party "Veo prompt" sites. Where Google's pages disagree, as on dialogue quotation marks, we report both with dates, and we found no published maximum prompt length. The templates are our own untested arrangements of the documented rules; Vinc makes an AI film studio.
AI Music for Film: Suno, Udio and Other Generators on Licences, Ownership, Lawsuits and Cue Sheets
Whether an AI-generated track can go in your film depends on the plan you made it on, whether the terms exclude film and TV, and whether anyone can own it at all. This guide compares the published terms of Suno, Udio, ElevenLabs Music, Stable Audio and AIVA, sets out the label lawsuits as reported, and explains what US copyright law and North American PRO rules mean for your cue sheet.