Veo 3, Google's AI for creating realistic videos with audio, is also available in Italy

Veo 3, Google 's artificial intelligence model that allows you to generate realistic videos with synchronized audio, is also available in Italy.

Users subscribed to Gemini Pro , a monthly plan that costs about 22 euros and allows access to the advanced features of Google's artificial intelligence, now have access to an enhanced version of the model, called " Veo 3 Fast ", also in our country. This variant is capable of generating videos with double the speed compared to the standard version. Videos created with Veo 3 Fast have a resolution of 720p and last a maximum of eight seconds.

“Pro” subscribers can make up to three videos. After this daily threshold, the application will automatically revert to using the previous model, Veo 2 .

What is Veo 3

Unlike the previous model - Veo 2 - and tools like OpenAI's Sora , Veo 3 doesn't stop at generating moving images. It integrates sound effects , ambient noises and even dialogues, synchronizing them perfectly with the action of the video. Google's AI is therefore able to produce videos with people speaking, in multiple languages ​​including Italian . The curious thing is that Veo 3 is even able to reproduce, more or less convincingly, the nuances of dialects.

Built on a diffusion-transformer architecture, Veo 3 leverages Google’s massive computational resources and likely a significant portion of YouTube ’s video archive for its training, though DeepMind – the team behind Big G’s most advanced AI – is being deliberately vague on the details.

The result is a model that can understand the laws of real-world physics and honor complex narrative prompts, generating surprisingly realistic 8-second clips.

A creative and technically advanced model

The Veo 3's ability to handle nuanced prompts—specifying camera angles, art styles, or even dialogue—clearly sets it apart from the competition.

For example, a prompt describing a sailor pointing to a stormy sea while reciting poetry generates a video with crashing waves, howling wind, and the sailor’s voice perfectly synchronized with his lip movements.

However, Veo 3 is not without its flaws : complex hand gestures, such as counting with fingers, can give it a hard time, likely due to poor representation in the training data.

But this model’s ability to realistically simulate physics — like an aurora borealis dancing across the Arctic sky — makes its videos look like real, professional footage.

Audio as a revolution

What makes Veo 3 truly revolutionary is its native fusion of video and audio , a technical achievement that has long been a challenge for AI developers.

Unlike rival models like Runway or Sora itself, which often require post-production audio syncing, the Veo 3 simultaneously generates both modes, ensuring perfect lip-sync and ambient sound coherence. This isn’t just a technological advancement: it’s a paradigm shift. One that can empower creativity but also give rise to controversial and offensive content.

The Danger of Deepfakes

To counteract any abuse, each frame is digitally marked with Google's SynthID watermark , invisible to the human eye but theoretically traceable. And, for a few weeks now, a small text, "Veo", has also been present at the bottom right of the generated videos.

However, this measure does not guarantee that less attentive users – for example, those who quickly scroll through content on social media – will realize that they are watching an AI-generated video.

AI-generated racist videos are circulating on TikTok that look like they were created with Veo 3 [the “Veo” in the bottom right leaves no doubt].

In the videos in question, black people are portrayed as apes or criminals. Google’s AI has also been used to construct false or misleading narratives about migrants.

These contents are garnering millions of views, despite spreading deeply offensive and prejudiced images.