Aiodic

Back to the table

97 Creative tool Vo Veo Creative Tools

Veo

Google's video model.

What it is

Veo is Google's family of AI video models. You describe a scene in words, such as "a lighthouse on a rocky coast at dusk, waves crashing, gulls calling", and Veo produces a short video clip of it that no camera ever filmed. You can also start from a photo, or give it several reference pictures of a character or object so they stay the same from shot to shot. Since Veo 3 in May 2025, the clips can come with sound made at the same time as the pictures: speech with moving lips, sound effects and background noise.

Veo is made by Google DeepMind, the company's AI research lab. Under the hood it is a diffusion model: it starts from random visual noise and cleans it up, step by step, into frames that match the prompt. Google's Veo 3 technical report says it works on a compressed version of the video and audio together, which is how the sound and pictures line up.

The current position, at the time of writing (September 2026). Google is moving its consumer video tools from Veo to a newer model called Gemini Omni, announced at Google I/O on 19 May 2026. Omni combines Gemini's understanding of text, images, sound and video with video generation, so you can make a clip and then edit it by talking to it ("make it night-time", "add rain"). Google's Gemini video page says Gemini Omni "will replace Veo in the Gemini app", and the Gemini app's help pages now describe Omni, not Veo, as the video model. Google made a developer version of Omni available on 30 June 2026 and released Gemini Omni 1.1 Flash on 27 August 2026. Veo has not disappeared, though:

So "Veo" today means two things: a line of models you can still pick by name in Flow and through Google's developer tools, and, according to press coverage of the I/O launch, part of the technology that Google's newer, all-in-one Omni model combines with Gemini. Model names and availability change fast, so check Google's own pages before relying on any of this.

How Veo got here

Why it matters

It made AI video talk. Earlier video generators, including the first versions of Sora and Runway, made silent clips; you added sound yourself. Veo 3 was one of the first widely available tools to produce dialogue, footsteps and room noise together with the picture. That one change turned AI clips from moving wallpaper into something that could pass for a news report, an advert or a scene from a film, which is why its launch drew so much attention.

It brought AI video to where people already are. Google put Veo inside the Gemini app, inside YouTube Shorts and inside Google's own video editor for businesses, Google Vids. For millions of people, their first AI video came from a button in an app they already used, not from a specialist website.

It raised the stakes for fake footage. Realistic video with realistic sound is exactly what a convincing hoax needs. Veo's history includes serious tests by journalists showing how easily it could fake news events, and Google's response with visible and invisible watermarks. Anyone who watches video online now needs the habits covered in the mistakes section below.

It shows how quickly the products shift. In about two years Veo went through five named versions and three apps, and at the time of writing its place in Google's own chatbot is being taken by a different model. For creators, the lesson is the same one the Sora shutdown taught: keep your own copies and your prompts, and don't build your whole workflow around one product name.

Real-world examples

Prompt Theory: the clip that went viral in two days (May 2025)

Two days after Veo 3 was announced, on 22 May 2025, the filmmaker and science communicator Hashem Al-Ghaili posted a short film called Prompt Theory (Made with Veo 3). Its idea was simple and unsettling: a series of AI-generated people, in street interviews and everyday scenes, insisting they are real and angrily denying that they were created by a prompt. Because Veo 3 made the voices, lip movements and background noise as well as the pictures, many viewers said they couldn't tell the characters apart from real people. It spread quickly across social media, and Android Authority included it in its round-up of the most striking early Veo 3 videos. Al-Ghaili made follow-ups in the same style. The film worked as both a showcase and a warning: the joke only lands because the fakes look and sound so real.

TIME makes fake news footage (June 2025)

On 3 June 2025, TIME published a test by reporters Andrew R. Chow and Billy Perrigo. Using Veo 3, they generated realistic clips of events that never happened, including a crowd setting fire to a Hindu temple in Pakistan, an election worker shredding ballots, and a car crash and arrest in Liverpool. TIME also reported what Veo refused: requests involving recognisable public figures such as Donald Trump and Elon Musk, and some false claims. Google told TIME it had "clear policies to protect users from harm". Google had announced on 30 May that it would add a small visible watermark to Veo 3 videos, with an exception for Ultra subscribers using Flow, and TIME noted the mark was small enough to crop out. Every Veo video also carries SynthID, Google's invisible watermark hidden in the pictures and sound. In December 2025 Google let anyone upload a video to the Gemini app and ask whether it contains SynthID. That check only finds Google's own watermark, so it can't tell you a video made by another company is real. The TIME test is still the clearest demonstration of why realistic AI video needs labels, and why labels alone are not enough.

A national TV advert made in two days (June 2025)

During Game 3 of the 2025 NBA Finals, the prediction-market company Kalshi aired an advert full of absurd scenes, including a farmer floating in a pool of eggs and an alien chugging beer, made with Veo 3. Its maker, the AI filmmaker PJ Accetturo, told Business Insider he produced it in two to three days for around 2,000 dollars. He used Gemini to turn a rough script into a shot list and prompts, then ran roughly 300 to 400 Veo generations to get 15 usable clips, which he edited together by hand. Coverage at the time contrasted this with the high cost of a traditional TV shoot. The example is useful for two reasons. It shows what a skilled person can now do cheaply, and it shows the hidden work: a script, hundreds of discarded attempts, and a human editor.

Disney's letter to Google (December 2025)

Veo has also been part of the argument over copyright. In December 2025 Disney sent Google a cease-and-desist letter, a formal demand to stop, accusing it of copyright infringement "on a massive scale" through its AI services and naming Veo among them, over videos showing characters such as those from Frozen and Star Wars. Days later, Engadget reported that Google had removed dozens of AI-generated videos of Disney characters from YouTube. The wider legal fight over AI and copyrighted characters, including Disney's lawsuit against an image company, is covered on the Midjourney page.

An everyday scenario: a book trailer on a budget

Picture Tom, a self-published author with a new thriller. He wants a 20-second trailer for social media. In Flow he uploads his cover art and a photo of a rain-soaked street as reference images, then asks for three short shots: a figure in a long coat walking away under a streetlight, a phone buzzing on a wet table, and a slow push-in on a door. His first attempts give the figure a different coat in every shot and a phone with gibberish on its screen, so he describes the coat in exactly the same words each time and asks for "no text on screen". He uses the free daily credits over a week, keeps the best four clips, adds his title and a line from a review in a normal video editor, and labels the post as AI-generated. Before running it as a paid advert, he checks Google's current terms and the ad platform's rules on AI content.

Try it yourself

Buttons, model names and free allowances change often. These steps use Google Flow, which at the time of writing offers both Veo and Gemini Omni and gives every signed-in user a small daily allowance.

  1. Open Flow. Go to flow.google.com (labs.google/fx/tools/flow redirects there) and sign in with a Google account. Google's Labs tools require you to be 18 or over and are available in a list of 100-plus countries.
  2. Check your credits. At the time of writing, Google's help page says everyone gets 50 Flow credits a day, with or without a subscription, and they don't roll over. A Veo 3.1 Lite clip cost 10 credits and a Gemini Omni clip 4 to 15, depending on length and quality. Without a subscription, video generation may be unavailable at busy times, roughly 2 PM to 5 PM UTC. Check the current figures on Google's credits page.
  3. Pick a model. Try Veo 3.1 Lite first, since it is the cheapest named Veo model. Clips are 4, 6 or 8 seconds long.
  4. Write a prompt that says who, where, what happens, the camera and the sound. Copy this one and change the details: > A slow push-in on an old wooden door in a stone cottage at night. Rain runs down a small window beside it. Warm candlelight flickers through the gap under the door. Sound: steady rain, a distant roll of thunder, then three slow knocks. No people, no text on screen.
  5. Generate and watch it twice, once for the picture and once with your eyes closed for the sound. Note what it got wrong: the number of knocks, the direction of the light, anything that melts or appears from nowhere.
  6. Change one thing at a time. Adjust a single part of the prompt, such as the camera movement, and generate again. Changing everything at once makes it hard to learn what each word does. This is iterative refinement applied to video.
  7. Try the same prompt with Gemini Omni in the same app and compare. Then try Omni's editing by asking for a change to the clip you already have, such as "make it snowing instead of raining".
  8. Download the clips you like, and save your prompts in a document. You'll want both if the product changes.
  9. Check before you publish. Label AI video as AI-generated, follow the platform's disclosure rules, and never make realistic video of a real person without their permission.

Common mistakes and misunderstandings

Thinking Veo is gone, or thinking nothing has changed. Neither is right at the time of writing. In the Gemini app, Google says Gemini Omni is replacing Veo. In Flow and in Google's developer tools, Veo 3.1 models are still listed. Guides written in 2025 that tell you to "open Gemini and pick Veo" may no longer match what you see.

Expecting a finished film from one prompt. Veo makes short clips, a few seconds each. Longer videos are built by extending shots or joining many clips in an editor. The Kalshi advert took hundreds of generations for 15 usable clips, and that was by a professional.

Assuming the watermark makes it safe to share, or that no watermark means it's real. The visible mark is small and some plans or tools haven't always added it. SynthID survives more editing, but only Google's tools can read it, and it proves nothing about videos from other companies. Content credentials based on the C2PA standard, a different kind of label explained on the OpenAI Images page, can be lost when a file is re-saved. Treat a missing label as "unknown", not "real".

Confusing "Google doesn't claim it" with "you own it outright". Google's Labs FAQ says Google won't claim ownership of what you generate, and points you to its terms for how you may use it. That is not the same as copyright protection, which in many countries doesn't cover material made by AI from prompts alone. The Runway page explains the US position. For commercial work, read Google's current terms for the exact product and plan you use, and ask a lawyer if the stakes are high.

Using real people, brands or characters. Google's rules forbid videos meant to deceive or harm, and the Disney letter shows that well-known characters can bring legal trouble. If you don't have permission, don't put it in the prompt.

Forgetting the costs. Free allowances are small and can pause at busy times. Serious use means a paid Google AI plan or paying per second through the developer tools, where Google charges by the second of video; it launched Veo 3.1 Lite in March 2026 at less than half the price of Veo 3.1 Fast. Prices change, so check them yourself.

Judging tools by the demos. Google, OpenAI, Runway and Kling all publish their best results and their own rankings. Try the same prompt in two or three tools and judge for your own project.

Related elements