Veo 3.1 Prompts: Structure, Audio, Shots and Examples to Copy

What Veo 3.1 is, and which version is current

Veo is Google's video-generation model, and Veo 3.1 is the current release in the Veo 3 family. It produces short clips with picture and sound together: dialogue, sound effects and ambience come from the model itself, so you don't have to add audio later. Checked in November 2026, there are three variants:

• Veo 3.1 (veo-3.1-generate-001): the standard model, and the only one that goes up to 4K.
• Veo 3.1 Fast: generally available since November 2025.
• Veo 3.1 Lite: in preview since April 2026; it doesn't support reference images.

Dates matter here. In the Gemini API, the Veo 3.1 preview models were shut down on 22 October 2026, and Google points developers to Gemini Omni Flash or to the Veo models on Vertex AI. Google's documentation also lists a retirement for Veo 3.1 Fast "on or after 17 November 2026", so it may stop being available soon. Before you start a project, check in your tool which version you have and whether it's still live; we make no assumptions about what the Gemini app or Flow offer today.

• Modes: text to video, image to video, first and last frame, reference images ("ingredients") and clip extension.
• Length: 4, 6 or 8 seconds per clip; 8 seconds only when you use reference images.
• Format: 16:9 or 9:16, at 720p or 1080p (4K on the standard model only), 24 frames per second.
• Extension: each step adds 7 seconds, up to 37 seconds in total.

How to structure a Veo 3.1 prompt

Google's official prompting guide uses a five-part formula, in this order: cinematography (shot, angle, movement, lens), subject, action, context (place, time, weather) and style and ambiance. On top of that you can add timing, audio and editing terms such as "match cut" or "jump cut".

Keep the prompt in English, which is the language listed on the model card. Leading with the camera tells Veo how to frame everything that follows.

[Shot type and camera movement] on [subject with 2-3 visible traits] as [one clear action] in [place, time of day, weather]. [Lighting], [style]. [Speaker] says, "[short line]." SFX: [one sound effect]. Ambient noise: [background sound].

Crane shot starting low on a lone hiker in a red jacket and rising high above her, revealing a mist-filled canyon at sunrise. Soft morning light, long shadows across the rocks, epic fantasy style. Ambient noise: wind moving through the canyon.

An 8-second clip has room for one action, maybe two. If your story has more beats, split it across several clips or several timestamped shots.

Audio in Veo 3.1: dialogue, sound effects and ambience

Veo 3.1 generates sound with the picture, but you get more reliable results when you ask for it explicitly, in sentences separate from the visual description. Google documents three patterns:

• Dialogue in quotation marks, with the speaker named: A woman says, "We have to leave now."
• Sound effects with an SFX: label: SFX: thunder cracks in the distance.
• Background sound with Ambient noise: Ambient noise: the quiet hum of a busy café.

A slow dolly-in on a chef plating a dessert in a busy restaurant kitchen, steam rising from the pans behind him. Warm tungsten light, shallow depth of field, cinematic style. The chef says, "Almost ready." SFX: a spoon taps the edge of the plate. Ambient noise: pans sizzling and muffled kitchen chatter.

Keep lines short enough to be spoken at a natural pace within the clip, and give each speaker a clear label ("the chef", "the older woman") rather than a pronoun.

Multiple shots, reference images and extensions

You can chain shots inside a single clip by starting each one with a timestamp. The timestamps should cover the whole length with no gaps, and each shot works best with one framing and one camera move.

[00:00-00:03] Medium shot of a lighthouse keeper in a navy wool coat climbing a spiral staircase, a lantern in his hand, warm light flickering on the stone walls. [00:03-00:06] Reverse shot from the top of the stairs as he reaches the lamp room and looks out at a stormy sea. Moody, cinematic style. SFX: boots on metal steps. Ambient noise: wind howling outside.

With reference images, open by saying what each image is for, so Veo knows which character, object or place to keep:

Using the provided images for the detective, the woman, and the office, a medium two-shot of the detective leaning across his desk as the woman slides a sealed envelope toward him. Late-night desk lamp light, film noir style. The woman says, "You didn't get this from me." Ambient noise: rain against the window.

To extend a clip, write the new prompt as a direct continuation: repeat the subject, wardrobe, lighting and style in the same words, and describe only what happens next.

Negatives: the negativePrompt field

Veo has a separate negativePrompt parameter for things you don't want to see. Google recommends writing it as a list of terms rather than instructions with "no" or "don't". Better still, describe what you do want in the main prompt: "empty white walls" works better than asking for no picture frames.

text, subtitles, watermark, logo, wall frames, extra fingers, cartoon

Camera and movement terms Veo understands

Veo understands plain film vocabulary. These are the terms that appear in Google's documentation; for other models, see the camera glossary.

• Shots: wide shot, close-up, extreme close-up, two-shot; low angle and high-angle.
• Moves: slow dolly-in, Truck left/right, Pan left/right, Tilt up/down, Pedestal up/down, Zoom in/out.
• Bigger moves: Crane shot, Arc shot or 180-degree arc, Aerial or drone shot, Tracking shot.
• Camera feel: Static (fixed), Handheld or shaky cam, Whip pan.
• Focus: deep focus or soft focus.

Some advanced angles and lens effects aren't officially supported. If Veo ignores a move, simplify it or fall back to a static camera.

Common Veo 3.1 mistakes

• Squeezing a long event into a 4 to 8 second clip: the model cuts it short, rushes it or leaves it unfinished.
• Asking for dramatic transformations (a car turning into a dragon): subtle, gradual changes work better.
• Writing negatives inside the prompt with "no" instead of using negativePrompt.
• Blending audio into the visual description, or not saying who speaks.
• Picking settings your version doesn't support, such as 4K on Fast or Lite, or 4 seconds with reference images.
• Naming real people, brands or copyrighted characters: apart from being a bad idea, safety filters may block the generation.

AI-generated video that could pass for real footage should be labeled, and each platform has its own rules about it. Check them before you publish.

Using the PromptNow Veo prompt generator

If you'd rather not assemble the prompt by hand, PromptNow's Veo 3.1 prompt generator does it for you. Pick the version, mode, number of shots, framing, camera move, lighting, style, length, aspect ratio, resolution and audio, then describe your scene. You get a prompt to paste into ChatGPT, Claude or Gemini, which returns the final Veo prompt with the right syntax, a negativePrompt list, two variants that each change one thing and a short note on what each part does.

To animate an image you already have, or to plan several clips before generating, there are also image-to-video and storyboard templates.