Veo 3.1 is the video model from Google DeepMind that generates audio together with the picture: dialogue, sound effects and ambience arrive in the same clip rather than being added afterwards. That changes how you write prompts, because you are now directing a scene with sound as well as a shot.
This guide covers what Veo 3.1 does well, how to structure a prompt, five example prompts you can adapt, the limits you will run into, and how to use it in Larund, the studio we build, alongside the other video models.
What Veo 3.1 is
Veo 3.1 is the third-generation line of Google DeepMind’s Veo video models. You give it a text prompt, optionally a reference image or a first frame, and it returns a short video clip. Its defining trait is native audio: speech, sound effects and background sound are generated in the same pass as the visuals, so lip movement and the sound of an action line up without a separate sound step.
Veo 3.1 also comes in lighter variants. In Larund you will see Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite. The Fast and Lite versions are the natural choice for drafts and for checking an idea before you spend more on the full model. As with every model, the exact price in credits is shown before you generate.
What it is good at
Veo 3.1 is known for strong prompt adherence, polished lighting and a cinematic look. It tends to follow a detailed description of a single scene closely, which makes it a good fit when you know exactly what the shot should contain.
- Scenes where sound is part of the shot: street ambience, rain, a door closing, a character speaking a short line
- Cinematic establishing shots, product shots and atmospheric b-roll with controlled lighting
- Short dialogue moments where the voice should match the mouth and the room
- Image-to-video, where you supply an approved frame and want it to move
How to structure a Veo 3.1 prompt
Think of the prompt as a shot description followed by a sound description. A reliable order is subject, action, setting, camera, lighting and style, then audio. Keep each clip to one idea; if you need two actions or two locations, make two clips and cut between them.
For speech, put the exact words in quotation marks and say who speaks and how. For sound effects and music, name them plainly and say whether they are loud or quiet in the mix. If you do not want a soundtrack, say so, otherwise the model may add one.
- Name the subject and what they are doing in one sentence.
- Describe the place and the time of day.
- Pick one camera move or shot size, such as a slow push-in or a locked-off wide.
- Describe the light and the overall look.
- Add the audio: dialogue in quotation marks, then effects and ambience.
Five example prompts
Adapt these by changing the subject and place. They are written to show the structure, not as guaranteed results; every generation varies.
Street at night: A woman in a yellow raincoat waits at a crossing in a quiet city street at night, rain falling, neon signs reflecting in puddles. Slow push-in from a medium shot, shallow depth of field, cool blue light with warm shop glow. Audio: steady rain, distant traffic, a tram bell once.
Dialogue: A chef in a small restaurant kitchen turns to the camera and says, "Tonight the soup is the best thing on the menu." Handheld medium close-up, warm tungsten light, steam rising. Audio: pans clinking, a low kitchen hum, no music.
Product: A glass bottle of cold brew coffee on a stone counter, condensation running down the glass, morning sun from the left. Locked-off macro shot, soft focus background, clean commercial look. Audio: a quiet ice clink and a soft room tone.
Landscape: A single hiker crosses a ridge above a sea of clouds at sunrise, wide aerial shot drifting forward. Audio: wind and distant birds, no music.
Animated style: A small paper boat floats across a puddle in a stylised watercolour world, camera low and level with the water. Audio: gentle water sounds and a soft music-box melody.
Limits to plan around
Clips are short, so a story is built from several clips rather than one generation. Characters can drift slightly between separate generations, which is why a locked reference image or a character sheet helps when a person must look the same across shots.
Spoken lines work best when they are short. Long speeches, very fast speech or several people talking at once are harder to get right, and it is normal to generate a few takes and keep the best. Text that appears in the picture, such as signs, can also need more than one attempt. Finally, the full model is among the more costly options per second, which is why drafting on Fast or Lite and then moving to the full model is a sensible habit.
Using Veo 3.1 in Larund
In Larund Studio you pick Veo 3.1 from the model list by its real name, see samples, and see the credit price before you generate. Larund’s video flow is keyframe-first: you approve a location sheet, a cast sheet and a start and an end keyframe, and only then is the motion rendered. That means you do not pay for motion on a shot whose look is wrong.
You can pin the framing with lens and camera-angle tokens in the prompt, such as &35mm or &low-angle. The same account and credit balance also cover Kling 3.0, Seedance 2.5, Hailuo H3 Max, Wan 3.0 and the other video models, so you can run the same shot on two models and keep the better one. The Free plan includes 5 credits a month plus 20 one-time welcome credits, with a small larund.com watermark on free videos; paid plans start at $9 a month and can turn the watermark off.
Is Larund the right pick?
A good fit when
- You want to try Veo 3.1 next to Kling, Seedance or Wan and compare results on the same shot
- You want to see the credit price before each generation and approve keyframes before paying for motion
- You want one balance and one account instead of a separate subscription per model
Probably not the best fit when
- You only ever use Veo and already have access through Google’s own tools
- You need a desktop video editor with a multi-track timeline; Larund generates clips and films rather than replacing an editor
- You need a long continuous take in one generation; clips are short and longer pieces are assembled from scenes
Frequently asked questions
What is Veo 3.1?
Veo 3.1 is Google DeepMind’s video generation model. It creates short clips from text or images and generates matching audio, including speech and sound effects, in the same pass.
Does Veo 3.1 generate sound?
Yes. Audio is generated with the picture, so you describe dialogue, sound effects and ambience in the prompt. If you do not want music, say so explicitly.
How do I write a good Veo 3.1 prompt?
Describe one subject, one action, the place, one camera move and the lighting, then add the audio. Put spoken words in quotation marks and keep each clip to a single idea.
Can I use Veo 3.1 without a separate Google subscription?
Yes. Larund runs Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite from one account and one credit balance, and shows the credit price before each generation.
Is there a cheaper way to test a Veo prompt?
Draft on Veo 3.1 Fast or Lite, or approve start and end keyframes first, and move to the full model only when the shot looks right.
Is there a free way to try Veo 3.1?
Larund’s Free plan gives 5 credits a month plus 20 one-time welcome credits, with a small watermark on free videos. How far that goes depends on the model and length you choose, and the price is shown before you generate.
Can Veo 3.1 keep the same character across scenes?
Separate generations can vary, so use a reference image or a character sheet. In Larund the cast sheet and keyframes you approve act as that reference.
Every top model. One studio.
Start free with Larund Studio: 25+ video and 30+ image models, keyframe films and motion graphics, with the price shown before every generation.
Try Larund free