Google Flow and Veo 3: Getting Started Guide

Flow is Google’s attempt to make video generation feel like making a film rather than filling in a form. It runs Google's Veo 3.1 models and Gemini Omni Flash, and wraps them in a project-based interface with a Scenebuilder for arranging clips into scenes. Veo is the headline: it generates video and sound together, including dialogue, effects and ambience, from text or a starting frame. This guide takes you through a first session. Plans, credit allowances and some labels change frequently, so where the interface differs from the description here, follow the interface.

What you need before you start

• A Google AI Plus, Pro or Ultra plan (or a qualifying Google Workspace plan), age verification at 18 or over, and a supported region. Flow uses AI credits that come with these plans; Ultra includes more credits and first access to experimental models. Check the current plan page for what is included, because the figures change.
• A clear idea of one shot. Flow projects can grow into sequences, but every sequence starts with a single clip of a few seconds.
• Optional reference images. If you want to animate a still or keep a character consistent, have the images ready. Flow’s own image tools are a convenient place to make them.

The three ways to generate

Flow’s prompt area has a mode switch. Which one you pick changes what the prompt should contain.

1. Text to Video. Describe the whole shot in words: subject, action, setting, camera, light, and sound. This is where to learn the model.
2. Frames to Video. Upload a start frame, and optionally an end frame, and describe the motion between them. The images decide the look; your prompt decides the movement.
3. Ingredients to Video. Upload reference images of characters, objects or styles and tell Flow how to combine them in a shot. This is the mode for keeping the same character or product across many clips, covered in Ingredients and Scenebuilder consistency.

Each generated clip lands in your project, where you can download it, extend it (8-second Veo clips) or add it to a scene in Scenebuilder. Note that the prompt box opens with the image model Nano Banana Pro selected: switch to Video before generating a clip.

Writing a first Veo 3 prompt

Veo 3 responds to plain, specific sentences. A useful order is subject and action, setting, camera, light, then a final sentence about sound. Because the model generates audio, the sound sentence is not optional padding: if you leave it out you get whatever ambience the model assumes.

A few habits that pay off from the first prompt:

• One subject, one action, one camera move per clip. Flow clips are short: 4, 6 or 8 seconds with Veo 3.1, and up to 10 seconds with Gemini Omni Flash.
• Use film vocabulary for the camera: close-up, wide shot, slow push in, tracking shot, handheld.
• Describe sound you can hear in a room: "rain on a tin roof, distant thunder, no music". Say "no music" if you plan to add your own.
• For speech, write the line in quotation marks and name who says it. Keep it under a dozen words. There is a full treatment in Veo 3 prompting for dialogue and sound.
• Avoid on-screen text, named brands and real people. They are unreliable, often declined, and you can describe the idea without them.

Settings at a glance

These are the controls you will meet in the current Flow interface. Labels and available values depend on your plan and may have moved by the time you read this.

A first session, step by step

1. Create a project and name it after the film or client, not the date. Everything you generate for that idea stays together.
2. Set Text to Video, 16:9, one output.
3. Write a 50 to 80 word prompt ending with a sound sentence, and generate.
4. Judge motion, framing and audio separately. If the picture is right and the sound is wrong, change only the sound sentence.
5. Fix one thing per regeneration. Note what you changed; credits disappear quickly when you alter three things at once and cannot tell which one helped.
6. Extend the clip once you like it: click it, choose Extend at the bottom and describe what happens next. Extension works on 8-second Veo 3.1 clips and is generated with Veo 3.1 Lite. Then add the clips to Scenebuilder to see them as a sequence.
7. Download finals as you approve them and keep the prompt with the file. A clip is hard to reproduce without its prompt, model and mode.

Credits are the main constraint, so treat the faster, cheaper model variant as your sketchbook and the full model as your final render.

Try it

A woman in a yellow raincoat waits at a rural bus stop as rain falls at dusk. Medium shot, camera static, shallow depth of field, headlights approaching softly out of focus behind her. Cool blue light with the warm glow of the bus stop lamp. Sound: steady rain on the shelter roof, a distant engine growing louder, no music.

For a second run, keep every sentence and change only the camera: "slow push in towards her face". For a third, change only the sound sentence. Two generations later you will know how much Veo 3 lets each part of the prompt control.