Veo (Google)

Veo is Google DeepMind's family of video generation models, which turn a text prompt or a set of reference images into short clips with synchronized sound. Introduced in May 2024, it became one of the best-known names in generative video, and Veo 3 (May 2025) was among the first major models to generate dialogue, sound effects and ambient audio together with the picture. As of October 2026 the current models are Veo 3.1 and its Fast and Lite variants; there is no Veo 4. Google is moving developers to a newer line, Gemini Omni, and the Veo 3.1 previews on the Gemini API shut down on October 22, 2026.

From Veo to Veo 3.1

Demis Hassabis, the chief executive of Google DeepMind, unveiled the first Veo at Google I/O on May 14, 2024, as a 1080p text-to-video model, a few months after OpenAI previewed Sora. Veo 2 followed in December 2024 with output of up to 4K and a better grasp of physics, although early public access was limited to short 720p clips in Google Labs. Veo 3, announced at I/O on May 20, 2025, ended what reviewers called AI video's silent era: it generated sound effects, background noise and character dialogue in sync with the action, and launched inside the Gemini app for subscribers to Google's top AI plan. Veo 3.1 arrived on October 15, 2025, and a cheaper Veo 3.1 Lite on March 31, 2026.

What Veo 3.1 Does

• Clip length. 4, 6 or 8 seconds per generation, with extension in 7-second steps up to 148 seconds at 720p.
• Resolution. 720p by default, with 1080p and 4K for 8-second clips. Lite stops at 1080p.
• Direction. A text prompt, up to three reference images, or a first and last frame for the model to interpolate between.
• Audio. Native sound on every clip.
• Provenance. Every output carries Google's invisible SynthID watermark.
• Price. On the Gemini API, Veo 3.1 costs $0.40 a second at 720p or 1080p and $0.60 at 4K; Veo 3.1 Fast starts at $0.10 a second and Lite at $0.05 (list prices as of October 2026).

Under the hood Veo belongs to the family of latent diffusion models that dominate text-to-video; Google has not published a detailed architecture paper for it.

Where It Runs, and the October 2026 Shutdown

Veo's status now depends on the platform. On the Gemini API, Veo 2 and the Veo 3.0 models closed on June 30, 2026, and the three Veo 3.1 preview models (standard, Fast and Lite) close on October 22, 2026, with gemini-omni-1.1-flash named as the replacement. On Google's enterprise platform (Gemini Enterprise Agent Platform, formerly Vertex AI), Veo 3.1 and Veo 3.1 Fast have been generally available since November 17, 2025, with retirement listed as "November 17, 2026 or later", and Lite is in preview there. Google's Vids app also uses Veo 3.1.

Veo also reaches users through other companies' products. Adobe offers Veo 3.1 among the Firefly partner models that Adobe Premiere and Firefly can call, and creative platforms such as Runway, Higgsfield and HeyGen list it in their model menus. For many creators, that is now how they meet Veo.

How It Compares

Veo is no longer at the top of the public rankings. On Artificial Analysis's blind-vote arenas as fetched on October 20, 2026, Veo 3.1 ranked 20th for text-to-video with audio and 14th for image-to-video, behind Gemini Omni 1.1 Flash (9th and 4th) and well behind the leaders, Alibaba's Wan 3.0 and ByteDance's Seedance 2.5. It keeps two real advantages over Omni: native 1080p and 4K generation, where Omni upscales anything above 720p, and longer single-shot extensions. Against standalone rivals such as Kling, its strengths are Google's enterprise terms and SynthID provenance, which helps with the platform disclosure rules covered in AI content labeling.

Veo's arc is also a lesson about building on video APIs. In 2026 OpenAI removed the Sora API entirely and Google retired Veo's previews in favor of a new model line. Teams that depend on a hosted video model should expect to migrate every year or so, and should keep prompts and pipelines portable across vendors.