Veo 3 is Google DeepMind’s video model that brought native sound to generated scenes. It combines visual action with dialogue, effects, and ambience, starting from a written brief or an image in supported workflows.
Google lists the original Veo 3.0 Vertex AI endpoints as discontinued and recommends Veo 3.1.
Native audio generation
Two ways to establish a scene
Standard and faster-generation variants
01 / Scene direction
Write the action you want to see.
Veo 3 turns a text description into a short scene. Specify who or what is in view, where the action happens, and how the camera follows it. A bookseller placing an ochre hardcover on a display table gives the shot a clear event; a slow move toward the cover gives that event a visual emphasis.
Name the subject, action, setting, and camera move in the same brief.
02 / A visual starting point
Give a still image its next moment.
Google introduced photo-to-video with Veo 3 in Gemini and Flow. The image establishes the person, setting, and composition; the accompanying brief describes what happens next. In this greenhouse scene, focus the movement on the watering can, the stream of water, and the fern leaves.
Keep the requested movement connected to objects already visible in the starting image.
03 / Sound inside the scene
Describe what the viewer hears as well as sees.
Veo 3 generates audio alongside video, including speech, environmental sound, and action effects. Give each sound a source: a conductor’s line, rain on the platform, or the train’s low idle. Short dialogue leaves room for the characters’ gestures and the surrounding atmosphere.
A single speaker and a short line make the intended exchange easy to follow.
Google’s original Vertex AI Veo 3.0 model specifications describe the following formats.
Duration
4 / 6 / 8 secondsShort clips with a defined action.
Orientation
16:9 or 9:16Landscape and portrait compositions.
Resolution
720p / 1080pDocumented Veo 3.0 output sizes.
Frame rate
24 fpsA fixed frame cadence for the shot.
Both belong to the Veo 3.0 family. Fast was introduced for quicker iteration, while the original model served the standard generation route.
| Dimension | Veo 3 | Veo 3 Fast |
|---|---|---|
| Text generation | Written scene brief | Written scene brief |
| Audio | Native sound generation | Native sound generation |
| Documented clip lengths | 4, 6, or 8 seconds | 4, 6, or 8 seconds |
| Documented resolutions | 720p / 1080p | 720p / 1080p |
| Workflow emphasis | Standard generation | Faster iteration |
| Current Vertex AI replacement | Veo 3.1 | Veo 3.1 Fast |
Explore Pexo workflows for product stories, cinematic moments, and moving photo sequences.
A focused scene is easier to direct when every instruction has a visible or audible role.
01
A quiet entrance hall, an oak console, a blue ceramic dish, and a woman holding a brass house key in the morning light.
02
She places the key gently into the dish. The camera moves closer and settles on her hand and the ceramic rim.
03
A small metallic clink as the key touches the dish, the rustle of her sleeve, and quiet room ambience. Leave a short pause after the sound.
Veo 3 is Google DeepMind’s video generation model introduced in May 2025. Its defining addition was native audio: it could generate dialogue, effects, and background sound together with the video.
Yes. Google introduced Veo 3 photo-to-video in Gemini and Flow in July 2025. A starting image provides the visual composition, while a written description directs the movement and sound.
The original Vertex AI Veo 3.0 specifications list 4-, 6-, and 8-second clips. Gemini’s Veo 3 photo-to-video launch described eight-second videos with sound.
Yes. Veo 3 supports native dialogue, environmental sound, and effects. Describe the speaker and words for dialogue, and connect other sounds to the objects or actions in the scene.
No. They are different model generations. Google lists the original Veo 3.0 Vertex AI endpoints as discontinued and points developers to Veo 3.1 or Veo 3.1 Fast.
A video model generates individual clips from inputs. A video agent works across the broader project, helping organize a brief, scenes, assets, revisions, and the finished edit. Choose according to whether you need one shot or a connected video workflow.
Veo 3 is Google DeepMind’s video generation model introduced in May 2025. Its defining addition was native audio: it could generate dialogue, effects, and background sound together with the video.
Yes. Google introduced Veo 3 photo-to-video in Gemini and Flow in July 2025. A starting image provides the visual composition, while a written description directs the movement and sound.
The original Vertex AI Veo 3.0 specifications list 4-, 6-, and 8-second clips. Gemini’s Veo 3 photo-to-video launch described eight-second videos with sound.
Yes. Veo 3 supports native dialogue, environmental sound, and effects. Describe the speaker and words for dialogue, and connect other sounds to the objects or actions in the scene.
No. They are different model generations. Google lists the original Veo 3.0 Vertex AI endpoints as discontinued and points developers to Veo 3.1 or Veo 3.1 Fast.
A video model generates individual clips from inputs. A video agent works across the broader project, helping organize a brief, scenes, assets, revisions, and the finished edit. Choose according to whether you need one shot or a connected video workflow.
