Google expands Veo 3 video model to 71 more countries

AI video generators used to produce silent, dreamlike clips. Veo 3 just gave them a voice, and it is eerily good at writing its own script.

About 100 hours after its debut at Google I/O, Google VP of Gemini Josh Woodward announced that Veo 3 is expanding access to 71 new countries. The model is rolling out across Google's consumer ecosystem, landing on Gemini on the web with mobile support coming soon, as well as Google's dedicated filmmaking tool, Flow.

Audio that writes itself

The headline feature of Veo 3 is native audio generation. Instead of spitting out a mute clip that requires third-party sound effects and voiceovers, Veo 3 synthesizes ambient noise, sound effects, and spoken dialogue in a single go.

In fact, it often talks even when you don't ask it to. Tests from The Verge revealed that simple prompts without any dialogue still resulted in characters speaking context-appropriate lines. A prompt featuring police officers generated a voice saying, "We need to clear the street," while a clip of cartoon cats on a dock featured them complaining that the fish weren't biting.

The realism is already turning heads. Alejandra Caraballo from Harvard Law School's Cyberlaw Clinic generated a convincing clip of a news anchor announcing the death of US Secretary of Defense Pete Hegseth, who is very much alive. A Reddit thread showcasing AI characters protesting their own creation prompts racked up over 50,000 upvotes, complete with hospital monitor beeps and tense standoff audio.

Tiers, quotas, and fuzzy guardrails

Google is rolling out Veo 3 through a clear subscription ladder. On Gemini, users can simply click the Video chip in the prompt bar. Pro subscribers receive a 10-pack trial, while Ultra subscribers get maximum limits with daily refreshes. On Flow, Pro users get 10 generations per month, while Ultra allowances jumped from 83 to 125 monthly generations.

Safety guardrails are noticeably active, but selective. Prompts asking for President Biden tripping or an anchor declaring a presidential assassination get blocked immediately. The model also refused to create deepfakes using uploaded personal photos paired with specific dialogue.

However, generic chaos still slips through easily. Testers managed to create realistic footage of Seattle's Space Needle on fire, or turn a personal photo of Mount Rainier into an erupting volcano with smoke and lava, paired with fake breaking news commentary. Another current limitation: extending clips currently forces a downgrade to Veo 2, which strips the audio entirely.

Why it matters

Veo 3 proves that the next battleground for generative video isn't resolution, it is audio integration. By bundling ambient sounds and synchronized speech directly into the prompt pipeline, Google has eliminated the tedious secondary step of sound design.

For solo creators and animators, this is an incredible workflow shortcut. But as tests demonstrate, it is also an unprecedented engine for high-velocity slop. With convincing audio generated in minutes, churning out hypnotic, low-effort children's content or viral panic bait for social platforms becomes trivial.

OpenAI has kept Sora under tight wraps, but Google seems eager to ship Veo 3 directly into paid tiers to drive Gemini subscriptions. It is a smart commercial land grab, though leaving the model free to invent its own realistic dialogue shows just how thin the line is between creative magic and convincing misinformation.