AI video24 min read5 October 2026

Veo 3 Guide for Marketers: Versions, Prompts and an Ad Workflow That Ships (2026)

What Veo 3 Is (and What Changed With Veo 3.1)

Veo is Google DeepMind's family of video generation models. Veo 3 arrived in 2025 and made one big change to AI video: it generates audio together with the picture. Before that, AI clips were silent and you had to add every sound later. With Veo 3, a prompt like "a barista says 'your flat white is ready' while the steam wand hisses" comes back with the line spoken, lips roughly in sync and the café sounds underneath.

Veo 3.1 followed as the production version. According to Google's Veo 3.1 model documentation, the 3.1 models support text to video, image to video, video from a start and end frame, extending a clip, and reference images to keep subjects consistent. They also attach Content Credentials, which record that a clip was made with AI.

What Veo Does Well

Native sound. Dialogue, sound effects and ambient noise come out of the same generation, which saves a lot of editing on short social clips.
Realistic motion and physics. Liquids pour, fabric moves and hair catches wind in a way that holds up on a phone screen.
Camera language. It understands terms like dolly in, tracking shot, low angle and shallow depth of field, so you can direct it like a camera operator.
Prompt following. It sticks to detailed instructions better than older models, which matters when a client has specific requirements.

What Veo Does Not Do

Long takes. Each generation is 8 seconds at most. Longer videos are built by extending clips or by editing several shots together.
Perfect text and logos. Words on packaging, signs and screens can still come out garbled. Plan to add text and logos in the edit.
Guaranteed consistency. Without reference images, the same "woman in a blue blazer" will look like a different person in every shot.
Truthful proof. Veo can show a product working. It cannot show your product working. That difference matters for advertising rules, and we come back to it below.

Veo 3 vs Veo 3.1 vs Fast vs Lite: Which Version to Use

Google now offers several Veo variants, and the names confuse almost everyone. Here is how they compare, based on Google's published specifications at the time of writing. Model names and limits change often, so check the documentation before you plan a big project.

Veo 3: Clip length: Up to 8 seconds; Aspect ratios: 16:9, 9:16; Output resolution: 720p, 1080p; Best for: Older projects and tools that have not moved to 3.1
Veo 3.1: Clip length: 4, 6 or 8 seconds; Aspect ratios: 16:9, 9:16; Output resolution: 720p, 1080p, 4K; Best for: Final renders, hero shots, anything that will carry ad spend
Veo 3.1 Fast: Clip length: 4, 6 or 8 seconds; Aspect ratios: 16:9, 9:16; Output resolution: 720p, 1080p; Best for: Testing prompts, storyboards, volume variants
Veo 3.1 Lite: Clip length: 4, 6 or 8 seconds; Aspect ratios: 16:9, 9:16; Output resolution: 720p, 1080p; Best for: High-volume drafts and simple shots (a preview model at the time of writing)

All 3.1 variants output 24 frames per second and can return up to four videos per prompt. Reference images (Google calls them asset images or "ingredients") work on Veo 3.1 and 3.1 Fast but not on Lite, and reference-image generations run at 8 seconds.

Our Rule of Thumb

We write and test prompts on Fast, then render the shots that made the edit on the full Veo 3.1 model. Fast runs quicker and uses fewer credits, so you can try ten variations of a hook instead of three. Once the camera move, the action and the timing are right, the full model adds detail in skin, fabric, reflections and sound that you notice on a big phone screen.

Use 4K only when you need it: a hero shot for a website, a cinema or outdoor screen, or footage you plan to crop. For Reels, TikTok and Shorts, 1080p vertical is what the platforms show anyway.

How to Get Access to Veo 3

There are several ways in. They all use the same model family, but they differ a lot in control, team workflow and where your files end up.

Gemini app: Who it suits: Solo marketers, quick ideas; Strengths: Simplest way to try Veo, works on mobile; Limits: Little control over settings, one clip at a time
Flow: Who it suits: Creators and in-house content teams; Strengths: Built for filmmaking: scenes, ingredients, start and end frames, extend; Limits: Needs a paid Google AI plan, less suited to automation
YouTube Shorts: Who it suits: Channel owners; Strengths: Generate clips inside the Shorts creation tools; Limits: Short, simple clips, limited control
Gemini API: Who it suits: Developers, apps; Strengths: Programmatic access, batch workflows; Limits: Needs coding, usage is billed per generation
Google Cloud: Who it suits: Agencies and enterprises; Strengths: Full settings, team permissions, data controls, 4K; Limits: Most setup, requires a cloud account and quotas

Gemini App

Open Gemini, choose the video option and type your prompt. This is the fastest way to see what Veo can do. Videos made in the consumer app may carry a visible watermark depending on your plan, and you cannot fine-tune settings like seed or negative prompt. Treat it as a sketchbook.

Flow

Flow is Google's AI filmmaking tool, built around Veo. It is where most creators should work. You can turn a text prompt into a clip, animate a still image, set the start and end frames, save characters and objects as ingredients, and extend a shot to build a longer scene. Flow also keeps your clips organized by project, which matters once you have fifty variants.

Gemini API and Google Cloud

The API and Google Cloud give you every setting: resolution, aspect ratio, duration, number of outputs, seed, negative prompt and reference images. This is how agencies and product teams run Veo at volume and plug it into their own tools. Google's cloud platform also adds team access controls and data residency options, which some regulated brands require.

Plans and Availability

Access depends on your Google AI plan and your country, and both change often. We do not list subscription prices here because they move, and we want this guide to stay accurate. Check Google's current plan page in your region before you commit. If you are in the UAE, confirm availability in your account, because some features reach the US before other markets.

Prompting Veo 3: The Structure That Works

Veo rewards prompts written like a shot description from a director, not a wish. A vague prompt such as "a woman using our app" gives Veo hundreds of decisions to make, and it makes them generically. A specific prompt makes those decisions for it.

Google's own Veo prompt guide lists the elements that matter. We use the same structure in this order, because it mirrors how a shot is planned on a real set.

The Seven-Part Prompt

Shot and camera. Framing, angle and movement: "close-up, eye level, slow dolly in."
Subject. Who or what, with specific detail: age range, clothing, hair, expression, product color and material.
Action. What happens in the 8 seconds, in order. One main action per shot.
Setting. Where it happens: room, time of day, props, background activity.
Light and color. "Soft window light from the left, warm tones" or "high-key studio light, white background."
Style. "Handheld phone footage, natural," "cinematic, shot on 35mm film" or "clean commercial look."
Audio. Dialogue in quotes, sound effects, ambience and music, or "no music."

Here is the difference in practice. A weak prompt:

A woman drinking coffee in a café.

A strong prompt:

Close-up at eye level, slow dolly in. A woman in her late twenties with curly dark hair and a mustard knit sweater holds a ceramic cup with both hands and smiles after a sip. Small café in the morning, wooden counter, espresso machine softly out of focus behind her. Warm window light from the right, gentle steam rising from the cup. Natural handheld documentary style. Audio: quiet café chatter, a cup set down on a saucer, no music.

The second prompt takes forty seconds longer to write and usually saves several regenerations.

Camera Moves That Veo Handles Well

Slow dolly in: What it does: Moves toward the subject; Use it for: Building attention, reveals, emotional lines
Tracking shot: What it does: Follows a moving subject; Use it for: Energy, walking-and-talking, sport
Orbit: What it does: Circles the subject; Use it for: Product hero shots, 360-degree reveals
Static tripod: What it does: No movement; Use it for: Talking heads, demos, authority
Handheld: What it does: Slight natural shake; Use it for: UGC style, realism, behind the scenes
Overhead top-down: What it does: Looks straight down; Use it for: Food, unboxing, desk setups
Crane up: What it does: Rises above the scene; Use it for: Establishing shots, endings

Use one move per shot. "Dolly in while orbiting and tilting up" usually produces a confused camera.

Directing Sound and Dialogue

Audio is where Veo 3 is most different from older models, and where most guides say the least. A few rules make a big difference:

Put spoken lines in quotation marks and say who speaks: The man looks at the camera and says: "This took me two minutes."
Keep dialogue short. In 8 seconds a person can say about 15 to 20 words naturally. Longer lines get rushed or cut off.
Describe the voice if it matters: "calm, warm voice," "excited, fast," "slight British accent."
List sound effects separately: "a zipper opening, then a soft thud as the bag hits the bed."
Say "no music" if you plan to add your own track in the edit. Generated music under dialogue is hard to remove later.
Ask for no on-screen text or subtitles. Veo sometimes adds captions on its own, and they are rarely spelled right. You will add proper captions in the edit.

Negative Prompts and What to Exclude

When you use the API or cloud tools, you can add a negative prompt listing what you do not want: "text, subtitles, logos, extra fingers, distorted face, cartoon style." In the apps, describe the result you want positively instead. Writing "an empty street at dawn" works better than "a street with no people," because naming the thing you want to avoid can pull it into the shot.

How We Turn Veo 3 Clips Into Ads: A 7-Step Workflow

A single Veo clip is a shot, not an ad. This is the workflow our AI video production team uses to go from a brief to a set of ad variants that are ready to test. It works whether you use Flow, the API or cloud tools.

Step 1: Write a One-Page Brief

Before you open Veo, write down:

the product and the one angle this ad will sell (speed, taste, status, or a benefit like "lasts all day")
the audience and where they will see it (Reels, TikTok, Shorts, YouTube in-stream)
the offer and the call to action
what must look exact: the product, the logo, the packaging, a presenter
what Veo must not show: claims you cannot prove, competitors' products, sensitive settings

This page stops you from generating beautiful clips that do not sell anything.

Step 2: Script the Hook, Body and CTA

Write a 15 to 30-second script in three parts. The hook carries the opening one to three seconds and decides whether anyone keeps watching. Write three different hooks for the same body, because the hook is what you will test. Our guide to AI video ads for Meta and TikTok covers the hook, body and CTA framework in more depth.

Step 3: Break the Script Into 8-Second Shots

Veo thinks in shots of up to 8 seconds, so plan in shots. A 24-second ad might look like this:

1: Time: 0-3 s; What we see: Hook: problem moment, close on the face; Camera: Handheld close-up; Audio: Line: "My skin hated the summer in Dubai."
2: Time: 3-8 s; What we see: Product picked up from a bathroom shelf; Camera: Slow dolly in; Audio: Soft bathroom ambience
3: Time: 8-14 s; What we see: Product applied, texture visible; Camera: Macro close-up; Audio: Gentle pump sound
4: Time: 14-20 s; What we see: Same woman outdoors, relaxed, golden hour; Camera: Tracking shot; Audio: Street ambience, light breeze
5: Time: 20-24 s; What we see: Product hero with offer card (added in edit); Camera: Slow orbit; Audio: Music in edit

Each row becomes one prompt. Shots shorter than 8 seconds can be generated at 4 or 6 seconds or trimmed in the edit.

Step 4: Lock Your References

If the same person or product appears in more than one shot, set up references before you generate anything. Use clean product photos and a reference image of your presenter as ingredients, or create a key still for each shot and animate it with image to video. We cover this in detail in the next section, because consistency is the most common reason Veo ads fall apart.

Step 5: Generate, Compare and Select

Generate each shot on Fast, two to four variations at a time. Pick the best take of each shot, adjust the prompt for the weak ones and regenerate. When the full sequence works as a rough cut, re-render the chosen shots on the full Veo 3.1 model at final resolution, using the same prompt and seed where your tool allows it.

Expect to throw away a good share of what you generate. Some shots land in two tries, others take ten or more. Budget time for that.

Step 6: Edit, Caption and Version

Bring the shots into a normal video editor. Add:

captions that are readable with the sound off (most feed viewing starts muted)
the logo, product name and offer as real graphics, never generated text
your music track and a voiceover if the ad needs one
an end card with the call to action

Then make versions: three hooks, two lengths (for example 15 and 30 seconds) and the aspect ratios you need. Veo outputs 9:16 and 16:9. For 4:5 and 1:1 feed placements, crop from 9:16 and keep the subject in the center safe zone.

Step 7: Launch as a Test

Run the variants as a test rather than picking a favorite. Watch hook rate (the share of viewers still watching after three seconds), hold rate, click-through rate and cost per result. Kill weak hooks early and feed the winners new variants. AI production makes this loop cheap in time, which is the real advantage over a single expensive shoot.

Keeping Characters and Products Consistent

Consistency is the difference between an AI clip and an ad a brand can sign off. Veo gives you four tools for it.

Reference Images (Ingredients)

You can give Veo 3.1 up to three reference images of a person, a character or a product, and it will keep them recognizable in the new shot. For products, use sharp photos on a plain background from more than one angle. For people, use a clear, well-lit photo of the face and the outfit. Reference-image generations are 8 seconds long.

Start and End Frames

Give Veo the opening image and the closing image of a shot, and it generates the motion between them. This is the most reliable way to control a product reveal or a transformation, for example a closed box at the start and the product standing next to the open box at the end.

Image to Video From Key Stills

For hero shots we often create the still frame in an image model, check the product against the real one, fix any errors, and only then animate it. A correct still is much easier to fix than a wrong video.

Extend

Extend continues a clip from its last second, so you can build a longer continuous shot. It works well for slow, simple scenes. Each extension adds some risk of drift in faces and details, so check every join.

Copy-Paste Character Descriptions

Even with references, write the subject description once and paste it word for word into every prompt: "Layla, early thirties, shoulder-length black hair, cream linen shirt, small gold hoop earrings." Changing even a word between shots invites Veo to change the person.

Veo 3 Prompt Templates for Common Ad Formats

These templates follow the seven-part structure. Replace the capitalized parts in square brackets with your own details. They are starting points, so expect to adjust them after the opening round of generations.

UGC-Style Testimonial

Handheld selfie-style close-up, vertical framing. [PERSON DESCRIPTION] sits in [HOME SETTING], holding [PRODUCT] near the camera, and says in a natural, friendly voice: "[ONE SENTENCE, UNDER 20 WORDS]." Soft daylight from a window, slightly imperfect framing, authentic phone footage. Audio: the line, light room tone, no music. No on-screen text.

Use this for honest feature explanations, not personal results. An AI person has never used your product, so it should not claim a result it could not have had. For real testimonials, work with real people. Our guide on what a UGC creator does explains how to brief them.

Product Hero Shot

Slow orbit around [PRODUCT, COLOR AND MATERIAL] standing on [SURFACE], studio setting with [BACKGROUND COLOR] backdrop. High-key soft lighting with a gentle rim light, crisp reflections, shallow depth of field. Clean premium commercial look. Audio: a soft whoosh as the camera moves, no music. No text, no logos.

Generate this from a product reference image or a checked key still. Add the real logo in the edit.

Problem and Solution

Close-up, static camera. [PERSON] in [SETTING] looks frustrated as [PROBLEM MOMENT]. They sigh. Cool, flat lighting. Natural style. Audio: [RELEVANT SOUND], a short sigh, no music.

Pair it with a second shot of the solution in warmer light. The contrast in light does a lot of the storytelling.

Founder or Expert Message

Waist-up shot, static tripod, eye level. [PERSON DESCRIPTION] stands in [OFFICE OR STUDIO], looks into the lens and says calmly: "[LINE]." Soft key light from the left, blurred background with plants. Clean corporate documentary style. Audio: the line, quiet room tone, no music.

Use AI presenters for explainers and offers. If the message is from a real founder, film the real founder or get their explicit consent for any likeness.

Seasonal GCC Spot

Wide shot, slow crane up. A family gathers around a table set for iftar on a balcony overlooking the Dubai skyline at dusk, lanterns glowing, dates and laban on the table. Warm golden light fading to blue hour. Cinematic, respectful, calm. Audio: soft conversation, distant city ambience, no music.

Seasonal spots need cultural care. Have someone from the audience you are targeting review every shot for dress, food, gestures and setting.

Troubleshooting Common Veo Problems

Garbled text or fake logos: Likely cause: Veo cannot spell reliably; Fix: Ask for "no text," add text and logos in the edit
Subtitles you did not ask for: Likely cause: Dialogue triggers captions; Fix: Add "no subtitles, no on-screen text," or use a negative prompt
Person looks different in each shot: Likely cause: No references, changing descriptions; Fix: Use reference images and paste identical descriptions
Product shape or label drifts: Likely cause: Model fills in missing detail; Fix: Use several product photos or a checked key still with image to video
Dialogue cut off or rushed: Likely cause: Line too long for the clip; Fix: Keep lines under about 20 words, or split them across shots
Odd hands or extra fingers: Likely cause: Complex hand action; Fix: Simplify the action, frame hands out, or regenerate
Camera does something random: Likely cause: Too many moves in one prompt; Fix: One camera move per shot
Generic stock-footage look: Likely cause: Vague prompt; Fix: Add specific details on light, style and setting
Clip refused: Likely cause: Safety filters (real people, brands, sensitive content); Fix: Remove real names, logos and sensitive elements, then rephrase

When a shot keeps failing after five or six tries, stop and change the approach. Split the shot into two simpler shots, switch to image to video from a still, or film that moment for real.

Using Veo 3 in Dubai and the GCC

Most Veo guides are written for a US audience. If you sell in the UAE, Saudi Arabia or the wider Gulf, a few things change.

Arabic and Bilingual Versions

Google's cloud documentation lists English as the prompt language for Veo 3.1, so write your prompts in English even when the ad is for an Arabic audience. Spoken Arabic dialogue from the model is worth testing, but check it with a native speaker every time: dialect, pronunciation and lip sync can all be off. For most client work we generate the visuals with natural ambient sound, then record or produce the Arabic and English voiceovers separately and cut them to the same shots. One set of visuals, two or more language versions.

Captions need the same care. Arabic captions run right to left and need a font that renders properly in your editor. Check the line breaks on a real phone.

Local Settings and Cultural Fit

Veo has seen far more footage of New York kitchens than of Dubai villas, Abu Dhabi offices or Riyadh cafés. If you want a Gulf setting, describe it in detail: architecture, materials, light, clothing and props. Review dress, gestures and food against your audience's norms before anything goes live.

Seasonal Timing

Ramadan, Eid, the UAE National Day, the Dubai Shopping Festival, White Friday and the back-to-school period all need fresh creative. Veo's speed makes it realistic to prepare seasonal variants of a proven ad instead of starting from zero each time. Plan the shots a month ahead, because review and approvals still take human time.

Advertising Rules

If you run influencer-style or presenter-led ads in the UAE, check the current UAE Media Council rules on advertiser permits and disclosure. AI presenters do not remove those obligations. Our guide to AI video production in Dubai covers the local process and how to choose a production partner.

Disclosure, Watermarks and Commercial Use

SynthID and Content Credentials

Google adds an invisible SynthID watermark to content made with its generative models, including Veo, so AI-made video can be identified later. The 3.1 models also support Content Credentials (C2PA), metadata that records how a file was made. Do not try to strip either. Platforms increasingly read these signals.

Platform AI Labels

TikTok, YouTube and Meta all have rules for realistic AI-generated content. YouTube, for example, asks creators to disclose altered or synthetic content when it looks realistic. Some rules apply to ads as well as organic posts, and the details change often, so check each platform's current policy at launch.

What AI Should Never Show

Labels do not make a misleading ad acceptable. Do not use Veo to show a result your product does not deliver, a before-and-after that never happened, a real person who did not agree to appear, or a customer review that no customer gave. These break advertising rules on every major platform, with or without an AI label.

Commercial Use

Google's terms for paid plans and its cloud services allow commercial use of what you generate, subject to their usage policies. Read the terms for the exact product and plan you use, and keep a record of prompts, references and outputs for every ad. If you use a person's likeness, a song or another brand's design as a reference, you need the rights to it regardless of what the AI terms say.

Veo 3 Quality Checklist Before You Publish

Run every clip and every final cut through this list. It takes ten minutes and catches most of what gets ads rejected or makes a brand look careless.

The product matches the real one: shape, color, label, cap, proportions.
No generated text, fake logos or garbled words anywhere in the frame, including the background.
Faces stay the same person across all shots.
Hands look natural and hold objects correctly.
Physics hold up: liquids pour the right way, objects do not merge or float.
Dialogue is clear, matches the script and the lips roughly sync.
No unwanted music, subtitles or watermark in the edit.
Captions are spelled correctly and readable on a small phone screen.
Claims are true and approved, with no invented results or reviews.
Platform disclosure is set where required.
Arabic or other language versions were checked by a native speaker.
Safe zones are clear: the product and captions are not covered by platform buttons.

Where Veo 3 Falls Short (and When to Shoot for Real)

We use Veo heavily, and we still shoot real footage for a good part of our work. Here is where it loses.

Products That Must Be Shown Working

If the sale depends on proof, such as a blender crushing ice, a stain disappearing or a phone case surviving a drop, AI footage is not proof. Viewers know it, and platform rules may treat it as misleading. Film the real product doing the real thing, then use Veo around it for scenes, b-roll and variants.

Exact Brand Assets

Complex packaging, small print, precise logos and specific architecture drift between shots. You can fix much of this with references and editing, but for a flagship brand film, a real shoot or CGI built from the product's 3D files may be faster and safer.

Real People and Testimonials

Authentic creator content still wins on trust for many categories, especially beauty, health and food. A real person with a real result beats a synthetic presenter for that job. Many of our best-performing campaigns mix both: real creators for proof, AI for volume and localization.

Long, Continuous Scenes

Anything that needs a single take longer than a few extended clips, with complex blocking and several people interacting, is still hard. Break it into shots or film it.

When to Bring in a Team

If you need many variants every week, several languages, brand-safe output and someone accountable for quality, a production team that combines AI tools with editors, scriptwriters and media buyers will usually get to a winning ad faster than a marketer working alone in Flow. That is exactly what our team does: we combine AI speed with human creative judgment and go from brief to launch in 7 days. You can see our AI video portfolio for examples, or compare Veo with other tools in our test of the best AI video generators for ads.

Frequently Asked Questions

What is Veo 3 and what can it do?

Veo 3 is Google DeepMind's AI video model. It turns a text prompt or an image into a short clip with sound, including dialogue, sound effects and ambience generated together with the picture. It handles realistic motion, lighting and camera moves well. Each generation is up to 8 seconds long, and longer videos are built by extending clips or editing several shots together.

What is the difference between Veo 3 and Veo 3.1?

Veo 3.1 is the newer version. It adds more control: start and end frames, extending clips, reference images to keep people and products consistent, and 4K output on the standard model. It also comes in Fast and Lite variants for quicker, lower-cost drafts. For new projects, most teams use Veo 3.1 Fast to test prompts and the full Veo 3.1 model for final shots.

How long can a Veo 3 video be?

A single Veo 3.1 generation is 4, 6 or 8 seconds long. To make a longer video you can extend a clip, which continues it from its last moment, or generate several shots and edit them together. Most social ads are 15 to 30 seconds, so they are usually built from three to six separate shots, each planned and prompted on its own.

How do I write a good Veo 3 prompt?

Write it like a director describing one shot. Cover the camera framing and movement, the subject in specific detail, one main action, the setting, the lighting, the visual style and the audio, with dialogue in quotation marks. Keep spoken lines under about 20 words, ask for no on-screen text, and use one camera move per shot. Test on the Fast model, then refine.

Can I use Veo 3 videos for ads and commercial work?

Generally yes. Google's terms for its paid plans and cloud services allow commercial use of generated content, subject to their usage policies, so check the terms for your exact plan. You still need rights to any likeness, music or design you use as a reference, you must follow each ad platform's AI disclosure rules, and you must never use AI to show results your product does not deliver.

Does Veo 3 work in Arabic and in the UAE?

You can use Veo from the UAE where your Google plan or cloud account supports it, so check availability in your account. Google's cloud documentation lists English as the prompt language, so write prompts in English. Test any spoken Arabic carefully with a native speaker. Many teams, including ours, generate the visuals and add Arabic and English voiceovers and captions in the edit.

Ready to Turn Veo 3 Into Ads That Sell?

Veo 3 makes great shots. Turning them into ads that convert takes a brief, a script, consistent references, careful QA, honest claims and a testing plan, and for GCC brands, proper Arabic and English versions. That is the part most teams find slow.

If you want that done for you, our team at XMA combines Veo and other AI tools with experienced editors, scriptwriters and media buyers, tested on 500+ paid campaigns across the US, UK and UAE. Tell us about your product and we will plan the shots, produce the variants and get them ready to launch in 7 days. Book a strategy call and get a quote for your next campaign.