HomeFull speed ahead. Keep track of advancements in every frontier
Genie world modelsGoogle DeepMind—VeoGoogle DeepMind—Google DeepMind—GeminiGoogle DeepMind—Tensor Processing Unit (TPU)Google DeepMind—AlphaFoldGoogle DeepMind—Genie world modelsGoogle DeepMind—VeoGoogle DeepMind—GeminiGoogle DeepMind—Tensor Processing Unit (TPU)Google DeepMind—GeminiGoogle DeepMind—
Veo
Courtesy of Google DeepMind · Press kit, editorial use (opens deepmind.google)

Veo

by Google DeepMindGBUS

DeployStage 4 of 5

Veo 3.1 (October 2025) is the current model, used in the Gemini app and the Flow video tool.

Updated 19 May 2026Checked 25 Sep0 updates this week

Milestones

No announced next step
  1. Veo announced at Google I/OMay 2024Complete.
  2. Veo 2 with 4K output and better physicsDec 2024Complete.
  3. Veo 3 adds matching sound. Flow video tool launchedMay 2025Complete.
  4. Veo 3.1 released15 Oct 2025Complete.

Most important updates

  • 19 May 2026
  • 15 Oct 2025
  • 20 May 2025
  • 16 Dec 2024
  • 14 May 2024

Current obstacles

  • Physics and consistencyClips can break physics or change a character's look between shots. Longer videos make this worse.
  • Deepfakes and misuseRealistic fake video can mislead people. Google adds hidden SynthID watermarks, but detection isn't perfect.

Physics limits

  • Video is enormous dataOne second of 1080p at 24 fps is about 150 million colour values. Models first compress it many times over; fine detail lost in that step can't be restored.
  • Long clips cost far more than short onesThe whole clip is denoised together and attention cost rises with the square of its length, which is why clips are around 8 seconds, not minutes.
  • Physics is imitated, not computedMotion is learned from how video usually looks. Nothing enforces gravity or solid objects, so hands, liquids and collisions still go wrong.

How it works

4 parts
Google's data centre in Council Bluffs, Iowa: Veo generates video on TPUs housed in campuses like this
Google's data centre in Council Bluffs, Iowa: Veo generates video on TPUs housed in campuses like thisPhoto: Chad Davis · CC BY 2.0 (opens commons.wikimedia.org)
Compress

Shrinking the video first

An encoder compresses video and sound into a much smaller code; the model works in that code and a decoder turns it back into pixels.

Diffusion

Carving from noise

Generation starts from random noise and removes it step by step, guided by the text prompt, until a coherent clip emerges.

Audio

Sound made alongside

Veo 3 generates dialogue, effects and music together with the frames, so lips and footsteps line up with the picture.

Marking

Invisible watermark

Every clip carries SynthID, a hidden watermark that Google's detector can find even after moderate edits or compression.

Update log

5 updates

Tue 19 May

  • Major: BlogSoftware

Wed 15 Oct 2025

  • Minor: BlogSoftware

Tue 20 May 2025

  • Major: BlogSoftware

Mon 16 Dec 2024

  • Minor: BlogSoftware

Tue 14 May 2024

  • Major: BlogSoftware

About Google DeepMind

The team behind Veo

Google DeepMind

Gemini, AlphaFold, Genie, Veo, TPU chips

Google DeepMind is Alphabet's AI lab, formed in 2023 from DeepMind and Google Brain. It builds the Gemini models and leads in AI for science. AlphaFold won a share of the 2024 Nobel Prize in Chemistry.

  • Gemini 3 (Nov 2025) led many benchmarks. Gemini 3.5 Pro, announced in May 2026, is delayed.
  • AlphaFold predicted structures for over 200 million proteins and won a share of the 2024 Chemistry Nobel.
  • Google's own Ironwood TPU chip is used in-house and sold to outside labs, including Anthropic.
Founded
201016 yrs
Headquarters
United KingdomUnited States
Status
Subsidiaryof Alphabet
Valuation
Alphabet-owned
Works in
AIFrontier modelsComputingBiotech
Coverage
5 programs · 31 updateslatest 3d agochecked 25 Sep
Customers
Anthropic
Partners
Isomorphic LabsBroadcom
Acquired by
Alphabet
People
Demis HassabisCEO and co-founderShane LeggCo-founder and Chief AGI ScientistJohn JumperAlphaFold lead; 2024 Nobel laureate
Primer