Prompt Veo for text-to-video

Veo generates a short clip from a written prompt, with sound and spoken dialogue made in the same pass. Pick it when there is no frame yet and the shot needs a character to speak: a trailer beat, an NPC greeting, a cutscene line.

Let the agent write it

You don't have to write any of this yourself. The Layer agent knows how to prompt each model it runs, from the same guidance this page is built on. Name the model and describe the shot, for example: Make a 4-second trailer shot with Veo 3.1 Lite: an orc warlord on a burning rampart at night, slow push in, and he growls "The walls fall tonight". With sound, no music. The agent writes the prompt, shows you its plan, and runs it when you approve. Use the rest of this guide when you want to write or adjust the prompt yourself, and see Write a brief for the agent for how to brief it.

When to use it

Use Veo for text-to-video when you need a shot from nothing and the sound matters: trailer and teaser beats, character barks and NPC greetings, cutscene previz, and social clips in landscape or portrait. It follows film vocabulary closely and speaks a written line word for word, with the character's mouth moving.

When the look is already approved as a still, start from that image instead, so the clip keeps it: see Prompt Seedance for image-to-video. For the basics of the video tool, see Create a video.

Versions on Layer

Veo 3.1

The full model, up to 4K. Render the keeper here once the prompt is settled.

Veo 3.1 Fast

The same inputs as Veo 3.1, also up to 4K, and quicker. A good middle step when you want to check a shot at higher quality before the final render.

Veo 3.1 Lite

The same prompts at up to 1080p, without 4K. Every test on this page ran on it: use it to find the shot, the timing and the line, then move to Veo 3.1. All three take 4, 6 or 8 seconds, a first and a last frame, and native audio.

How to write the prompt

Shot and camera, then subject, action, setting, style, then the spoken line and the sound:

Low-angle medium shot, slow dolly in. A battle-scarred orc warlord in heavy iron armour stands on a burning castle rampart at night, embers swirling around him. He raises his war axe and looks straight into the camera. Stylised 3D game cinematic, warm firelight against a deep blue night sky. The orc growls: The walls fall tonight. Sound: crackling fire, distant war drums and wind. No music.

Google's own formula puts cinematography first, then subject, action, context, and style and ambiance; you don't need every part in every prompt. With no image to start from, the prompt has to set the whole scene, so name the subject and setting plainly and give the sound its own sentences at the end. That prompt is the one in the clip above.

What works

Write the line after a colon

Name who speaks and how, then a colon and the line: "The orc growls: The walls fall tonight." Google advises against quotation marks, which can make the model print the line on screen. We ran the same shot both ways on the same seed, and again on a portrait tavern shot with a longer line: no captions appeared in any of the four clips, and a transcript of each read the line word for word. Quotes are safe in practice; the colon costs nothing, so it is the form we use. Keep the line short: a 9-word greeting fits a 4-second clip.

Sound in its own sentences

With Generate audio on, describe the sound after the picture, in separate sentences: sound effects ("SFX: thunder cracks in the distance"), ambience ("Ambient noise: low tavern chatter and a crackling hearth") and any line of dialogue. Say "No music" when the clip will sit under your own score.

Film terms for the camera

Veo follows standard cinematography vocabulary: low angle, over-the-shoulder, dolly in, truck left, crane up, arc shot, whip pan, static shot. Name the shot size and one move. In our tests "low-angle medium shot, slow dolly in" gave exactly that framing and a slow push.

One moment per clip

A 4 to 8 second clip holds one focused moment. "The detective finds a clue, then drives across the city, then confronts a suspect" comes out muddled; generate each as its own clip and cut them together in the video editor.

Several shots with timestamps

When one clip needs a cut, give each shot a time range: "[00:00-00:04] Wide shot of the knight cresting the hill. [00:04-00:08] Close-up of her face as she sees the dragon."

JSON prompts

Many community guides recommend writing Veo prompts as JSON. We ran the trailer shot above as a JSON object on the same seed:

{
  "shot": "low-angle medium shot",
  "camera": "slow dolly in",
  "subject": "a battle-scarred orc warlord in heavy iron armour",
  "setting": "a burning castle rampart at night, embers swirling around him",
  "action": "he raises his war axe and looks straight into the camera",
  "style": "stylised 3D game cinematic, warm firelight against a deep blue night sky",
  "dialogue": { "speaker": "the orc", "delivery": "growls", "line": "The walls fall tonight." },
  "audio": ["crackling fire", "distant war drums", "wind", "no music"]
}

Both versions gave the same shot, the same push in, and the line spoken word for word. They differed only the way two runs do: the JSON orc raised his axe a moment later. JSON is a way to organise a prompt, not a way to improve it: the model reads the keys as text, so none of them is a setting, and Duration, Generate audio and Aspect ratio still come from the form. Google's own guides are written in prose. JSON earns its place when a script or workflow assembles prompts from parts.

Name the style

Say what the clip should look like: "stylised 3D game cinematic", "cel-shaded animation", "claymation", "Japanese anime style". Google calls style one of the strongest levers on the result.

Keep a character across clips

Give the character a name, a fixed description (build, age, hair, face, costume) and a voice ("a deep, gravelly voice"), and paste that text unchanged into every prompt. Keep the same Seed too. It makes results more alike, but it does not guarantee the same clip.

What to avoid

"No" in the negative prompt

The Negative prompt field takes the unwanted things as plain nouns: "urban background, man-made structures". Writing "no buildings" or "don't show buildings" there works against you. In the main prompt, describe the absence: "a desolate landscape with no buildings or roads".

Settings that matter

Generate audio is off by default; turn it on for dialogue and sound. Duration takes 4, 6 or 8 seconds; find the shot at 4. Aspect ratio sets the shape and resolution together: 16:9 or 9:16, at 720p, 1080p and, on Veo 3.1 and Veo 3.1 Fast, 4K. Among the advanced settings, Negative prompt takes what to keep out, and Seed makes reruns more alike. First frame and Last frame turn the job into image-to-video. Above the prompt, Camera, Staging, Look and Performance take reference sets for each.

Prompts to start from

Trailer shot with a spoken line

Low-angle medium shot, slow dolly in. A battle-scarred orc warlord in heavy iron armour stands on a burning castle rampart at night, embers swirling around him. He raises his war axe and looks straight into the camera. Stylised 3D game cinematic, warm firelight against a deep blue night sky. The orc growls: The walls fall tonight. Sound: crackling fire, distant war drums and wind. No music.

NPC greeting for a vertical clip

Set Aspect ratio to 9:16.

Medium close-up, static camera. A dwarf innkeeper with a braided red beard leans over a wooden bar in a warm, candlelit tavern, polishing a tankard. He looks up at the camera and says: Welcome, traveller. The ale is on the house tonight. Stylised 3D game cinematic. Sound: low tavern chatter and a crackling hearth. No music.

Use it in Layer

1. In a session, open Create video from the side panel.

2. Click the model name at the top of the panel and choose Veo 3.1 Lite while you find the shot.

3. In the prompt box, write the shot and camera, the subject and action, the style, then the spoken line and the sound.

4. Turn on Generate audio. Leave Duration at 4 seconds while you iterate.

5. Click Create video. The cost shows on the button before you click.

6. When the clip lands on the board, click View to open it in the viewer and play it with sound. Change one thing in the prompt at a time; when the shot is right, render it on Veo 3.1.

Learn more

• The ultimate prompting guide for Veo 3.1 (Google Cloud): the five-part prompt formula, camera and lens vocabulary, audio cues for dialogue, sound effects and ambience, and timestamped multi-shot prompts.
• Video generation prompt guide (Google Cloud): every prompt element with worked examples, the full list of shot types and camera moves, and how to word a negative prompt.
• Best practices for generating videos (Google Cloud): one moment per short clip, writing speech without quotation marks, and keeping a character and voice consistent across shots.
• Generate videos with Veo 3.1 (Google AI for Developers): what each Veo 3.1 version supports, and what the seed does and does not guarantee.
• We tested JSON prompting in Veo 3 (VP Land): a side-by-side of JSON and prose prompts for the same scene, concluding that JSON organises a prompt rather than improving it.

Related

• Create a video
• Prompt Seedance for image-to-video
• Choose a model
• Write a brief for the agent