Veo 3: how to make AI video with Google in 2026

I remember the first time I saw a Veo clip with sound. An old fisherman on a pier turns to the camera and says something about the weather, while seagulls cry and water splashes behind him. Eight seconds. Nobody filmed him, nobody recorded a voiceover, nobody edited anything: the clip came out of one paragraph of text.

Before Veo, video AI made silent movies. The picture looked great, and then you were on your own: hunting for music, picking sound effects, recording a voice, and syncing it to the lip movements by hand. An eight-second clip ate up an evening. Veo made sound part of the generation, so the clip comes out ready to post straight from your browser window.

I'll walk through Veo so you can make your first clip while you read. Inside: what the name Veo 3 actually means today, where to use it, how many clips you get for free, who can access it and how to pay, step-by-step instructions for video from text and from a photo, how to keep one character across several clips, how to write dialogue, a prompt formula with 12 ready prompts, 10 business scenarios, and a big troubleshooting section. Read it in order or jump to what you need from the table of contents.

Section 01What Veo 3 is in 2026

Veo is a family of video generation models from Google DeepMind. The first version came out in 2024, the second at the end of that same year, and in May 2025 Google showed Veo 3. It made a splash with one thing: the clip came out with sound already in it. Characters talked, dishes clinked, the wind blew, and all of it lined up with the picture in time.

In October 2025 the Veo 3.1 update arrived. The sound got richer, characters held their appearance better from shot to shot, and it learned to work with several reference images and with a first and last frame. Then on June 30, 2026, Google shut down the old Veo 3, Veo 3 Fast, and Veo 2. So when you search for "Veo 3" today, you effectively land on Veo 3.1. The old name lives on in search out of habit.

Veo 3.1 comes in three versions, and mixing them up gets expensive:

These prices are for regular accounts. For subscribers on the top Ultra plan, Lite and Fast cost half as much, while Quality stays the same.

What Veo can do in practice:

• make a clip from a text description;
• bring to life an image or photo you upload;
• build a scene between a first and a last frame you provide;
• keep the same character or object across different clips using reference images;
• generate sound: speech with lip sync, sound effects, background music;
• extend a finished clip by continuing the action from its last frame.

One generation lasts 4, 6, or 8 seconds. For 1080p and 4K, pick 8 seconds; the shorter options only come in base quality. Formats: horizontal 16:9 and vertical 9:16 for reels and stories.

Section 02Where Veo works: Flow, Gemini, and other entry points

There's no single website called Veo. The model is built into several Google products, and each has its own personality. Here's a map so you don't get lost.

For steady, ongoing work I recommend Flow. You can see how many credits you have left, choose a specific Veo version, assemble several clips into one scene, extend a clip, and download it in the quality you need. Gemini is simpler but gives you less control: you don't always know which model is running at the moment, and you can't steer length and quality as precisely.

Gemini is handy for something else: coming up with the idea. Ask it to describe five scenes for a reel about your service, pick the best one, and take it to Flow. I covered how to use Gemini itself in detail in How to use Gemini.

Section 03What Veo costs: free and paid plans

Everything in Flow runs on credits. Every account, including free ones, gets 50 credits a day. They expire by the next day, so you can't stockpile them. Google's paid plans add a monthly allowance on top.

Exact prices depend on your country and change over time, so check the Google One plans page before you pay. A rough money benchmark: on the Pro plan, one clip on Fast works out to about 40 cents, and one on Quality to about two dollars.

What you can realistically do for free: five test clips on Lite a day or two proper ones on Fast. That's enough to learn how to write prompts and to make a couple of reels a week. For regular ads and daily clips, the free allowance runs short, and the Pro plan is a sensible place to start.

A separate note for developers. Through the API and AI Studio, Veo is billed per second: Lite costs about 5–8 cents a second, Fast 10–12 cents, and the full version 40 cents at 1080p and 60 cents at 4K. For a regular user this route is pricier and more complicated than Flow; I mention it just to complete the picture.

Section 04Getting access: regions, age, account, and payment

For most people in the US, getting into Veo takes a minute. Still, there are a few conditions, and when Flow refuses to open, it's almost always one of them. Here's what to check.

Paying for a plan. The Google AI Plus, Pro, and Ultra plans are billed through Google One, the same subscription that gives you extra Google Drive storage. You pay with a regular bank card or through Google Play on Android. Some countries also offer a free trial month of Pro, and students sometimes get special offers; check what the Google One page shows for your account. After you subscribe, the monthly credits appear in Flow automatically.

Section 05The Flow interface in two minutes

Flow looks like an editing app that's been boiled down to a single window. No need to be intimidated: you only need five spots.

Projects. The home page shows your projects. It's convenient to keep one project per task: "reels for the week," "challenge ads," "lesson intros." All your clips and images pile up inside the project.

The prompt box. At the bottom of the screen there's a large input field. To its left is the mode switch: Text to Video, Frames to Video, Ingredients to Video. The mode determines what you attach to your description.

Generation settings. The slider icon next to the prompt box. This is where you choose the model (Lite, Fast, Quality), the aspect ratio (16:9 or 9:16), the length, and how many variations to make at once: one or two. Two variations cost twice as much, but for important clips it's handy to pick the better one.

The results feed. Finished clips appear above the prompt box. Hover over a clip and buttons appear: download, extend, use a frame as the start of the next shot, add to scene.

Scene. A separate mode where clips line up one after another on a track. This is where you assemble several eight-second pieces into a 30–60 second video and download it as a whole.

Flow adapts its interface language to your browser, and the buttons are few. You can write prompts in plain English, and English is also the language Veo understands best for camera moves and lighting.

Section 06Your first video from text, step by step

Let's start with a clip made entirely from a description. A task almost anyone can use: an atmospheric intro for a reel about how you start your workday. Eight seconds, vertical, with sound.

Why it worked. The first sentence sets the place and time. The character is described briefly: a sweater and a desk, no life story. The actions go in order and fit into eight seconds. A separate "Sound:" line tells Veo what to play, so the audio comes out precise, with no random music.

The most common beginner mistake is cramming a whole movie into eight seconds: "she wakes up, walks to the kitchen, makes coffee, sits down to work, calls a client." Veo will either skip half the actions or blend them into mush. One scene means one or two actions. If you need a story, make several clips and assemble them into a scene.

Section 07Video from a photo and from two frames

For business, this mode is often more useful than text. You start with your real photo, so the clip keeps your office, your product, your mug with your logo on it. In Flow it's called Frames to Video.

First and last frame. Upload both frames, and Veo builds the transition between them. It's a ready-made trick for before-and-after clips: the first frame is a messy room, the last is the same room after a cleaning service. Or an empty plate and a plate with the finished dish. In this case the prompt describes how the transformation unfolds: "the camera stays still, things fly back into place on their own, the light gets warmer."

For simple tasks like "animate an old family photo" or "turn a portrait into a short animation," there are simpler tools. I rounded them up in How to animate a photo with AI.

Section 08One character across several clips

The curse of every video AI is that the character changes from clip to clip. In the first clip the barista has a red beard, in the second it's light brown, and in the third his apron has vanished. For a series of reels or a three-scene ad, that's unacceptable. Flow has a mode for exactly this: Ingredients to Video.

An ingredient is a reference image: a character, an object, or a style. You upload up to three of them, and Veo builds the scene from them while keeping their look. For example, the ingredients could be a photo of your character, a photo of your branded mug, and an image of a café interior. The prompt describes what happens: "the character puts the mug on the counter and smiles at a guest."

How to make a series of three clips with one character:

Extending. If a good clip ends too soon, click Extend on it. Veo looks at the last frames and continues the action for a few more seconds. That way you can stretch a scene to a minute or longer. Each extension slightly raises the chance that the face or the background drifts, so for reels one or two steps is usually enough.

Section 09Sound and speech: how to make a character talk

Sound is the reason people choose Veo. It generates three layers: the characters' speech, sound effects (footsteps, clinking dishes, a car engine), and ambience (café buzz, ocean waves, music). All of it in sync with the picture.

How to write a line. Say who's speaking and put the text in quotes: The woman looks at the camera and says: "Three mistakes that cost you clients." Veo understands this is direct speech and syncs the lips. The line has to fit into 8 seconds including pauses: that's 10–15 words, no more. A long sentence gets cut off in the middle or comes out as a tongue twister.

Other languages. Veo speaks English most cleanly. It can also deliver lines in Spanish, French, German, and many other languages, but stress and pronunciation slip more often there, and unusual names and brand words may come out wrong even in English. What works: keep lines short and simple, without rare words; generate two variations and pick the better one; or give the character only the emotion and lay your own voice over it in editing.

How to describe sound effects. On a separate "Sound:" line, list what you hear in order of importance: Sound: rain on the window, a kettle starting to whistle, soft lo-fi music in the background. If you don't want music at all, say so: No music, only natural room sound. Otherwise Veo sometimes adds music to its own taste.

Section 10The prompt formula and 12 ready prompts

A video prompt describes what changes over time. In an image you capture a moment; in a clip you describe movement and sound. That's why the Veo formula has six blocks:

Three to five sentences are usually enough. Not every clip needs all the blocks: add dialogue only if the character has to talk. For camera movement, keep a few simple words handy: push-in (the camera moves closer), pull-out (moves away), orbit (circles around), tracking shot (follows the character), static (stays put), handheld (shot by hand, with a slight shake).

Below are 12 prompts for common tasks. Copy them and change the details to fit your niche.

Note the "Dramatized testimonial" prompt: use it only to act out a real review with the client's permission, and label it as a dramatization. Section 12 explains why. If you want more ready-made wording for captions, posts, and stories to go with these clips, take a look at the AI prompt library.

Section 11Let Claude write your Veo prompts

Writing prompts with the six-block formula is a skill that comes after about a week of practice. You can speed it up by handing the job to a text AI. I use Claude because it keeps the format tidy and doesn't pad the description.

Go to claude.ai and sign up with your email or Google account. The free plan is enough for writing prompts.

Then give Claude the instructions once, and after that just describe your tasks in plain words.

Then you paste the prompt into Flow. If the clip doesn't come out right, tell Claude what you didn't like: "the camera is jittery, make it steadier," "she smiles too widely." It rewrites the part that needs it. After a dozen clips you'll have your own set of phrasings that work; save them in your notes.

You can go further and hand Claude the whole reel script: the hook, three scenes, the lines, the captions. I showed how to do that for a whole month ahead in AI for reels: 30 scripts in one evening.

Section 12Where Veo clips pay off: 10 scenarios

A pretty clip made just to be pretty doesn't bring in money. Here are ten tasks where eight seconds from Veo genuinely help sell services, courses, and consulting.

The first two usually pay off fastest. The hook decides whether people finish the reel, and a live announcement brings people in better than an image with a date on it. Start with those.

And one observation from practice. Clips where a made-up character speaks work well for humor and metaphors. Where trust is at stake (testimonials, promised results, your personal experience), made-up people do damage. A testimonial from a client who doesn't exist is deception, and your audience spots it quickly. For trust, use your own face and real stories.

Section 13How not to burn your credits

Fifty credits a day feel generous right up to your first failed generation on Quality. Here are the rules that save both credits and nerves.

• Draft on Lite, final on Fast or Quality. Test the idea for 10 credits first. Once the composition and action look right, regenerate on a better model with the same prompt.
• One variation at a time. Two variations cost twice as much. Turn them on only for important clips where you want a choice.
• Quality only when you need 4K or a close-up of a face. For reels watched on a phone, Fast is almost indistinguishable.
• Change one detail. A failed clip is almost always ruined by one thing. Change everything at once, and you won't know what helped.
• Text first, then video. Ask Claude or Gemini to check the prompt: does the action fit into 8 seconds, are there any contradictions.
• Keep winning prompts in your notes. Your personal archive of phrasings is your biggest saving after a month.
• Spend your daily credits every day. They expire overnight. Two tries a day on free credits means 60 practice clips a month without paying anything.

Section 14When something goes wrong

Nearly every Veo problem happens to everyone. Here are the common cases and what to do about them.

Section 15Veo vs other AI video tools

Veo is strong at cinematic visuals and sound. But for some tasks its neighbors are more convenient. A short comparison table:

How to choose in ten seconds. You need a clip where the character talks, or an ad-quality shot: Veo. You need people from your own photos and a longer clip for less money: Kling. Many people keep two tools: Veo for clips with sound, Kling for animating photos. What happened to Sora and what replaced it, I covered in Sora shut down, and a general overview of tools for different jobs is in Best AI tools 2026.

Section 16Pre-publish checklist

Before you post a Veo clip, run through this list. One minute of checking saves you awkward comments.

• The format matches the platform: 9:16 for reels and stories, 16:9 for YouTube and your website.
• The character's face is the same from the first frame to the last.
• Hands and fingers have no extra joints.
• The speech matches the lips, and names and unusual words sound right.
• There's no garbled text or gibberish signs in the frame.
• No unwanted music, and the voice and background are balanced.
• The watermark is handled: covered with a caption bar or out of the way.
• The clip has no faces used without consent and no other brands' logos.
• If the clip is an ad, you're not passing off a made-up character as a real client.
• The first two seconds hook: movement, a spoken line, or an unexpected shot.

RecapThe whole path in a minute

Start today with one clip: take the hook of your next reel, ask Claude to turn it into a prompt with a spoken line, and generate it on your free credits. Put those eight seconds at the start of the video and check your completion rate a couple of days later. Chances are you'll want to make a second one.