Google is moving quickly to put Veo 3 in front of more people. About 100 hours after the AI video model’s initial launch, the company is expanding access through the Gemini app to users in 71 additional countries.
The rollout gives more Gemini subscribers a way to test AI-generated video that combines moving images and sound. It also makes the practical limits of Veo 3 clearer: who gets access, how many generations they can use, where it works, and what the model still cannot do.
Access is expanding, but not everywhere
The update came from Josh Woodward, Vice President of Gemini at Google, who announced the expansion on X. The key point is scale: Veo 3 is no longer limited to its earliest launch group, and Google is opening the model to a much broader set of users.
EU countries are not included in the rollout for now. The source does not give a reason for that exclusion, so the most important confirmed detail is simply that users in those countries remain outside this phase of availability.
For people who can access it, Veo 3 arrives through Gemini rather than as a separate public tool. That matters because the experience, limits, and subscription rules are tied to Gemini plans and to Flow mode, Google’s mode aimed specifically at AI filmmakers.
Subscription tiers define how much users can generate
Google is not giving every Gemini user the same amount of Veo 3 access. Gemini Pro subscribers receive a trial package of ten Veo 3 generations through the web interface. That package is a one-time trial offer.
Users with an Ultra subscription at $250 per month get the maximum number of generations Google allows, including daily refreshes. In Flow mode, the difference between plans is also clear: Ultra users get 125 generations monthly, while Pro subscribers receive ten generations per month.
Those limits shape how people are likely to use the model. A one-time trial is enough to test prompts and understand the basic output. A larger monthly allowance, especially with daily refreshes, is more suited to repeated experimentation, iteration, and longer creative workflows.
Veo 3 still has important restrictions
The broader rollout does not remove every constraint. For now, Veo 3 works only in the web version of Gemini Pro. That means the experience is not described as available across every Gemini surface in the source article.
Audio support is also limited. Veo 3 currently supports only English audio output, though other languages may occasionally surface. Flow mode has another restriction: it does not support voice output when users upload their own images.
These details are important because Veo 3’s main appeal is not just that it creates video. The attention around the model comes from the combination of visuals and audio, and from how closely it can follow prompts. Any limit on language, voice output, or where the feature works directly affects what users can produce.
Prompt control is driving the attention
Veo 3 is emerging as another viral AI product from Google after last year’s NotebookLM Audio Overviews. Even with limited access, users have been posting demo videos across social media that show how the model handles detailed scene requests.
The source describes the model as following prompts with impressive precision. One example involved a dachshund moving from a living room through an open front door and onto a porch, then watching an ice cream truck pass by. The point is not the dog itself, but the level of scene continuity users are now expecting from AI video tools.
That combination of precise prompting, video generation, and audio output is why Veo 3 is attracting so much attention. It suggests a future in which a short text instruction can produce a scene that feels complete enough to share, remix, or present as a finished clip.
The same realism raises trust problems
The source also makes clear that Veo 3’s strengths create risk. If a model can produce realistic video and sound from a simple prompt, it can also make fake material easier to create. One example cited in the source involved a fake interview on a fictional car show.
The same capability could be applied to fabricated protest footage or other misleading material. That is why Veo 3 adds to existing concerns about generative AI and disinformation. The issue is not only that fake content can be made, but that it can now be made with less effort and less technical skill than before.
A few years ago, creating a deepfake, even one focused on replacing someone’s face in a video, required hours of work and serious technical skills. According to the source, today a single line of text can generate realistic scenes with images and sound.
That shift changes what viewers need to assume when they encounter visual content from unverified sources. Veo 3 shows how quickly AI-generated video is improving, but it also makes scrutiny more important. As access expands, the central question is not only what users can create, but how audiences can judge what they are seeing and hearing.