Veo 2

Veo 2 is the release that made Google DeepMind the reference point for cinematic AI video: believable physics, stable long takes, and prompts that behave like camera direction. It is no longer the current model. The Veo page at DeepMind now leads with Veo 3.1, which returns eight-second clips at up to 1080p and 4K and folds native audio — dialogue, effects and ambience — into the same pass, reachable through the Gemini app, Flow, AI Studio and Vertex AI. Veo 2 itself was silent, so treat this entry as the start of the line, not the model to buy today.

The limits are the ones shared by every clip generator. Eight seconds is not a scene, so narrative work still means assembling shots in Flow or in a real editor, and cross-shot character consistency trails Runway and the multi-image reference workflow in Kling. Check before paying: which Veo version your plan routes to, since a listing can serve an older or a Fast variant; per-second API pricing, roughly USD 0.40 at standard quality and about 0.60 for 4K with audio, Fast between 0.10 and 0.30, with no free API tier; duration and resolution ceilings per tier; that SynthID marking is embedded in output; and whether commercial use rides on a consumer subscription or needs a cloud contract.