Veo

What Veo is

Veo is a generative video model from Google DeepMind. Google announced it at its I/O event on 14 May 2024 and described it as the company's most capable video model at that point^[1]. Like other generative video systems, it takes a written prompt and produces a moving image sequence rather than a single still picture; the model has to keep objects, motion and camera behaviour consistent across the whole clip.

How it was introduced

Veo was not released as an open tool. Google first offered it in a private preview to a limited set of creators through VideoFX, a product reached by joining a waitlist, and said some of its capabilities were planned for YouTube Shorts and other products^[1]. That pattern — a research announcement, a small invited preview, then gradual integration into existing products — is how the company introduced the model to the public.

Main characteristics

According to Google, Veo generates high-quality 1080p video in a wide range of cinematic and visual styles^[1]. 1080p refers to a common high-definition resolution used for online and broadcast video, so the claim puts the output in the same format range as ordinary production footage rather than small experimental clips. Google also said the model could generate videos that go beyond a minute in length^[1], longer than the few-second clips that early text-to-video systems typically produced.

Google said Veo understood cinematic terms such as "timelapse"^[1]. This matters because prompts written by people who make films tend to use the vocabulary of filmmaking — shot types, camera movement, pacing — rather than plain descriptions of objects. A model that responds to those terms is meant to be usable by someone who already thinks in shots.

Watermarking

Videos generated with Veo inside VideoFX were watermarked with SynthID^[1]. SynthID is Google's watermarking technology, which the company describes as able to embed imperceptible digital watermarks into AI-generated images, audio, text and video^[1]. The mark is designed not to be visible or audible to a viewer, but to be detectable by software, so that generated material can be identified afterwards. For a video model, this addresses a practical concern: once clips look like ordinary footage, viewers cannot easily tell how they were made.

Place in the line of models

Veo was succeeded by Veo 2 in December 2024^[2]. The first Veo therefore stands as the model that introduced the name and the generation of capabilities that later versions built on — 1080p output, minute-scale clips, response to filmmaking language, and watermarking of generated video when used in Google's creator products^[1].