To get a good Veo 3 video, write one shot in plain English in this order: camera, subject, action, setting, style, then the sound on separate lines: dialogue in quotes, effects after SFX:, background after Ambient noise:. That is Google’s own formula for Veo 3.1, the current version. A clip is 4, 6 or 8 seconds, 720p to 4K, 16:9 or 9:16, with generated sound.
Veo 3 is real and still in use: Google announced it on 20 May 2025 (not December 2024, as this page used to say; that was Veo 2). It was replaced by Veo 3.1 on 15 October 2025. Since May 2026 the Gemini app makes video with a newer model, Gemini Omni, so if you want Veo specifically, you now reach it through Flow, Google Cloud or a third-party service. The prompt advice below works for both.
Want to skip the typing? The Veo 3 prompt generator builds this structure from a few fields, free, in your browser.
Veo 3 in November 2026: what changed
| Date | What happened |
|---|---|
| 20 May 2025 | Veo 3 at Google I/O: the first Veo with native sound (dialogue, effects, ambience) |
| Jul 2025 | Veo 3 and Veo 3 Fast in the Gemini API and generally available on Vertex AI |
| Sep 2025 | 9:16 vertical and 1080p; API price cut from $0.75 to $0.40 per second (Fast: $0.15) |
| 15 Oct 2025 | Veo 3.1: up to 3 reference images, scene extension, first and last frame |
| Mar–Apr 2026 | Veo 3.1 Lite, the cheapest tier |
| 20 May 2026 | Gemini Omni Flash at I/O 2026; the Gemini app’s video help now names Omni, not Veo |
| Aug 2026 | Gemini and Flow let you turn off the visible watermark (SynthID stays) |
| 22 Oct 2026 | The Gemini API shut down its Veo 3.1 preview models |
There is no Veo 4. Pages promising one are guesses.
What Veo 3.1 can do
- Length: 4, 6 or 8 seconds per generation. Scene extension builds each new clip from the last second of the previous one, so a scene can run “a minute or more”, in Google’s words.
- Resolution: 720p and 1080p; 4K on full Veo 3.1 (Fast and Lite stop at 1080p). 24 fps.
- Aspect ratio: 16:9 or 9:16. Set it in the tool, not in the prompt.
- Sound: dialogue with lip sync, sound effects and ambience, generated with the picture.
- Inputs: text, a start image, a first and last frame (Veo fills the transition), or up to 3 reference images of a character, object or place.
- Language: prompts are officially English only. More on that below.
- Marking: every video carries an invisible SynthID watermark; you cannot remove it.
Google’s prompt formula for Veo 3.1
From Google Cloud’s Ultimate prompting guide for Veo 3.1:
[Cinematography] + [Subject] + [Action] + [Context] + [Style & Ambiance]
What each part means in practice:
- Cinematography: shot size and movement first. Medium two-shot, slow push-in. Crane shot rising. Handheld selfie at arm’s length.
- Subject: who or what, with visible detail. “A tired waitress in a mint 1970s uniform” beats “a woman”.
- Action: one main action per clip. Eight seconds is short.
- Context: where and when. Place, weather, time of day.
- Style and ambiance: light, lens, film stock, mood. Warm tungsten light, 35mm grain.
Sound gets its own sentences:
- Dialogue in quotes:
She says, "We have to leave now."A line has to fit in the clip: one or two short sentences for 8 seconds. - Sound effects:
SFX: thunder cracks in the distance. - Background:
Ambient noise: the quiet hum of a starship bridge. - Several shots in one clip: timestamp blocks,
[00:00-00:02] wide shot… [00:02-00:04] close-up…. - Leaving things out: describe what you want instead of writing “no”. “A desolate landscape with no buildings or roads” works; “no man-made structures” works worse. In the API’s negative prompt field, list the unwanted things without “no” or “don’t”.
Two habits from the community that Google never recommended: writing prompts as JSON and adding “no subtitles”. JSON helps only because it makes you fill every slot; the model reads it as text either way. “No subtitles” is a common fix for burned-in captions under dialogue; it is harmless, so our generator adds it when there is speech.
10 Veo 3 prompt examples
The first four are Google’s own examples, word for word. The rest we wrote with the same structure. Each opens in the generator for tweaking.
Cinematic reveal (Google)
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.
Build it in the Veo 3 prompt generator →
Two characters talking (Google)
Camping (Stop Motion): Camper: "I'm one with nature now!" Bear: "Nature would prefer some personal space."
Build it in the Veo 3 prompt generator →
ASMR sound (Google)
A keyboard whose keys are made of different types of candy. Typing makes sweet, crunchy sounds. Audio: Crunchy, sugary typing sounds, delighted giggles.
Build it in the Veo 3 prompt generator →
Negative prompt (Google’s pattern)
Prompt: A lone oak tree in an open meadow, leaves rustling, stylized animation. Negative prompt: urban background, man-made structures, dark, stormy, or threatening atmosphere
Build it in the Veo 3 prompt generator →
Dialogue scene
Medium two-shot, slow push-in. In a dim 1970s diner booth, a tired waitress in a mint uniform pours coffee for a nervous young man in a rain-soaked jacket. She says, "You've been sitting here three hours, hon." He replies, "I'm waiting for someone who isn't coming." Warm tungsten light, 35mm film grain. Ambient noise: rain on the window, a jukebox playing low.
Build it in the Veo 3 prompt generator →
“Prompt theory” street interview
The 2025 trend where generated people find out they are generated still works, and Veo’s lip sync is what makes it land.
Street-interview style, handheld medium close-up. A man in a grey suit holds a microphone toward a startled woman on a sunny city sidewalk. He asks, "What do you think about the rumour that we're all just AI prompts?" She laughs nervously and says, "That's ridiculous... right?" then glances up at the sky. Documentary look, natural daylight. Ambient noise: traffic, distant birds.
Build it in the Veo 3 prompt generator →
Vertical vlog (set 9:16 in the tool)
Handheld selfie-style vertical video, front-facing phone camera at arm's length. A cheerful woman in her late 20s with curly red hair and a yellow raincoat walks through a busy Tokyo street market at night and says to camera, "Okay, I found the best ramen in the city, and it's in a basement." Neon signs, light drizzle. Ambient noise: crowd chatter, sizzling grills.
Build it in the Veo 3 prompt generator →
Product shot
Slow 360-degree arc shot around a matte black wireless earbud case on a wet slate pedestal. The lid opens on its own and both earbuds rise a few millimetres with a soft blue glow. Studio lighting, rim light, reflections on the wet stone, premium commercial style. SFX: a soft magnetic click, a low synth swell.
Similar recipe in the library →
ASMR macro
Extreme close-up, static camera, shallow depth of field. A sharp knife slowly slices through a glass-like crystal apple on a dark walnut board, revealing a glowing amber interior. Soft top light, black background. SFX: crisp glassy crunch with each cut. No music, no voice.
Build it in the Veo 3 prompt generator →
First and last frame
Upload a first frame (a seed in an open palm) and a last frame (a grown oak at sunset), then:
Time-lapse: the seed in the open palm sprouts and grows into a towering oak as the camera slowly pulls back and rises. Golden-hour light. SFX: soft creaking wood, rising wind, a gentle orchestral swell.
Similar recipe in the library →
More shot recipes: camera movement vocabulary, UGC-style ad, seven-part prompt structure, character consistency.
Prompts in other languages
Veo 3.1 officially takes English prompts only. Gemini Omni says English is fully supported and other languages “may work with varying results”.
Speech in another language is a different matter: write the scene in English and quote the line in the target language, naming it.
Close-up, static camera. An elderly fisherman mending a net on a wooden pier looks up and says in Spanish: "El mar nunca miente." Overcast morning light. Ambient noise: gulls, waves against the pilings.
Build it in the Veo 3 prompt generator →
Google has not published anything on non-English dialogue. Community reports say it works, but less reliably than English: sometimes the line comes out in English or slurred, so plan for a few tries. If the words must be exact, generate in English and dub.
Our Veo 3 generator has the interface in Spanish, Russian and Turkish and a speech-language field, and it always writes the prompt itself in English.
Where to use Veo, and what it costs
- Gemini app: needs a Google AI Plus, Pro or Ultra plan. Since May 2026 the app’s video help describes Gemini Omni as its video model; Google publishes no daily video count, only multiples of a standard limit.
- Google Flow (labs.google/flow): Google’s filmmaking tool, with Veo 3.1 and Omni, paid in credits. A small number of free daily credits is reported by third parties; check inside Flow.
- Google Cloud (Vertex AI, now called Gemini Enterprise Agent Platform in the docs): Veo 3.1, 3.1 Fast and Lite, us-central1, on provisioned throughput rather than pay-as-you-go.
- Gemini API: the Veo 3.1 previews closed on 22 October 2026; the API’s video model is now Omni Flash, about $0.10 per second of 720p video.
- YouTube: Omni Flash is free inside Shorts Remix and the YouTube Create app, 18+.
Reported per-second prices for Veo 3.1, from The Decoder in March 2026 (Google’s own Cloud price for Veo 3.1 was not available to us):
| Model | 720p | 1080p | 4K |
|---|---|---|---|
| Veo 3.1 | $0.40 | $0.40 | $0.60 |
| Veo 3.1 Fast | $0.10 | $0.12 | $0.30 |
| Veo 3.1 Lite | $0.05 | $0.08 | n/a |
So one 8-second Veo 3.1 clip is about $3.20 at 1080p, or $0.40 on Lite at 720p.
How Veo compares now
On the Artificial Analysis text-to-video arena with audio (read 2 November 2026), Veo 3.1 sits at #22, Veo 3.1 Fast #23, Lite #24. Google’s Omni Flash 1.1 is #10. The top places belong to Wan 3.0, MiniMax H3 and Seedance 2.5. Veo 3 was the model to beat in mid-2025; in late 2026 it is a solid, well-documented option with 4K and long scene extension, not the leader.
OpenAI closed the Sora app in April 2026 and the API in September, so “Veo 3 vs Sora 2” is no longer a choice anyone has to make. For the other models see our guides to Hailuo, Seedance, Kling and Midjourney video.
Fixes for common problems
- Captions burned into the video: add “no subtitles” or describe the frame without text.
- Garbled text on signs: keep text out of the shot and add titles in an editor.
- The line gets cut off: shorten it. One or two sentences per 8-second clip.
- Two actions blur together: split them into timestamp blocks or two clips joined with scene extension.
- The character changes between clips: use reference images (up to 3) and repeat the same description word for word. See style lock.
FAQ
Does Google Veo 3 exist?
Yes. Google DeepMind announced Veo 3 on 20 May 2025 as its first video model with native sound. It has since been replaced by Veo 3.1 (October 2025), which is the version Google Cloud and Flow offer now.
How long can a Veo 3 video be?
Each generation is 4, 6 or 8 seconds. Veo 3.1 can extend a scene clip by clip to a minute or more; Gemini Omni in the Gemini app extends to 40 seconds.
Does Veo 3 support prompts in languages other than English?
Officially, Veo 3.1 prompts are English only. Dialogue in another language usually works if you write the scene in English and quote the line, for example: says in Spanish: “Hola”. Results are less reliable than English.
Does Veo 3 generate audio?
Yes. Veo 3 and 3.1 generate dialogue with lip sync, sound effects and ambient sound together with the picture. Put speech in quotes and label effects with SFX: and background with Ambient noise:.
Is Veo 3 free?
Not in the Gemini app: video needs a Google AI Plus, Pro or Ultra plan. Free ways to try Google’s video models are Omni Flash in YouTube Shorts Remix and YouTube Create (18+). The Gemini API has no free tier for video.
What are the best practices for prompting Veo 3?
Follow Google’s order: camera, subject, action, setting, style. One action per clip, specific visual detail, sound on separate lines, and describe what you want instead of what you don’t. Set aspect ratio and length in the tool rather than in the prompt.
Can I remove the Veo watermark?
The visible watermark can be switched off in Gemini and Flow settings since August 2026, except where the law requires it. The invisible SynthID watermark stays on every video.
Sources
Checked on 2 November 2026:
- Google: generative media at I/O 2025 (Veo 3 announcement)
- Google Developers: Veo 3.1 in the Gemini API
- Google Cloud docs: Veo 3.1 (lengths, resolutions, English-only prompts, region)
- Google Cloud: the ultimate prompting guide for Veo 3.1
- Google Cloud docs: Veo prompt guide
- Google DeepMind: Veo prompt guide and Veo model page
- Google Developers: Veo 3 pricing and 9:16/1080p
- Gemini API: video generation, Omni and pricing
- Google: I/O 2026 announcements (Gemini Omni)
- Gemini Apps Help: video generation and usage limits
- The Decoder: Veo 3.1 Lite prices
- Artificial Analysis: text-to-video leaderboard
- Gigazine: visible watermarks optional in Gemini