Veo

Veo is a family of generative artificial intelligence video models developed by Google DeepMind. First announced in May 2024, Veo generates high-definition video clips from text prompts, still images and reference material, and later versions produce synchronised audio — including dialogue, sound effects and ambient noise — together with the visuals. The models are available to developers through the Gemini API and Google Cloud's Vertex AI, and have powered video generation across Google products including the Gemini app and Flow; the family competes with systems such as OpenAI's Sora, Kuaishou's Kling AI and ByteDance's Seedance.^[1][2]

History

Google announced the first Veo model at its I/O developer conference on 14 May 2024, describing it as capable of generating 1080p videos more than a minute long with an improved understanding of physics and cinematic concepts. A second-generation model, Veo 2, followed in December 2024 with improved realism and motion quality, positioned as Google's answer to rival video generators of the period.^[4]

Veo 3, released in May 2025, was the first version to generate audio natively: dialogue, sound effects and ambient sound are produced in the same generation pass as the picture, a capability Google demonstrated at that year's I/O keynote. The release significantly raised expectations for audio-visual quality in AI-generated video and intensified competition with other video-generation services.^[3]

In October 2025 Google introduced Veo 3.1, adding richer audio, more narrative control and enhanced realism, and bringing audio to editing features such as ingredients-to-video, frames-to-video and scene extension. Veo 3.1 became generally available on Vertex AI on 17 November 2025, and a lower-cost Veo 3.1 Lite entered preview in April 2026. Google has continued to position the family as its leading video model while a new generation of models reached the market.^[2][5]

At I/O 2026 in May, Google announced Gemini Omni, a new multimodal family whose first member, Gemini Omni Flash, generates and edits video from combinations of text, image, video and audio inputs, and stated that Omni would replace Veo inside the Gemini app over time while Veo remains available to developers through the Gemini API and Vertex AI.^[6]

Key Concepts and Technology

Veo models generate video from natural-language descriptions and reference images, with later versions accepting video and audio as additional inputs. Outputs are typically eight-second clips at 720p, 1080p or 4K resolution that can be extended in successive generations or edited through tools built around the model, such as Google Flow, a filmmaking application that combines Veo with other generative models. Google has released limited detail about the model's internal architecture, describing it primarily through its capabilities: cinematic camera control, consistent characters and environments across shots, and an understanding of physical behaviour of objects and light.^[1][5]

The defining feature of the third generation is joint audio-video generation. Rather than adding a soundtrack afterwards, Veo 3 and later models synthesise speech, effects and ambience in coordination with the scene, enabling dialogue-driven clips that lip-sync to generated voices. On Google Cloud's Vertex AI, Veo 3.1 outputs can carry C2PA Content Credentials, an industry provenance standard that records how a piece of media was created, and businesses can access the models through the Gemini API and Vertex AI with enterprise controls.^[5]

Applications and Impact

Veo is used broadly for short-form social video, advertising and e-commerce content, product demonstrations, previsualisation of film and television scenes, and rapid creative prototyping, where generations that once required shoots or animation pipelines can be produced from a prompt. Google integrated generative video into its consumer products, including the Gemini app and YouTube Shorts, and in 2026 began offering access to its newest video model at no cost to Shorts and YouTube Create users.^[6]

The model family sits at the centre of a fast-moving market for generative video that expanded from 2024 to 2026, with Western and Chinese developers — OpenAI, Google, Kuaishou, ByteDance, Alibaba and others — releasing competing models in quick succession and driving down the cost of generated footage. As with other video generators, the technology raises concerns about deepfakes, misinformation and likeness rights, and provenance features such as C2PA credentials and platform labelling have become a focus for regulators and platforms seeking to distinguish synthetic media from recorded footage.^[5]

>See Also

• Sora
• Kling AI
• Seedance
• Gemini
• Google DeepMind
• AI Video Generation