Google launches new artificial intelligence Veo 2 that can create videos in up to 4K resolution

Google’s DeepMind announced its new video-generating artificial intelligence, Veo 2, this Monday (16). Can create new model clips longer than two minutes It reaches resolutions up to 4K (4096x2160p).

The new model offers “improved understanding” of real-world physics, a recurring flaw in Sora’s generations and Better understanding of the nuances of human movements and expressions.

According to Google, Veo 2 understands unique aspects of cinematographysuch as genres, camera lenses and cinema effects. The company talks about the ability to add 18mm camera effects; as well as specific capture angles.

Theoretically, Veo 2 can create videos in 4K resolution and up to 2 minutes in length. This makes the Google model 4 times superior in terms of pixel count and ensures that the generations last 6 times longer than OpenAI’s Sora.

Despite, This advantage is only theoretical: VideoFX, the testing tool that Veo 2 is currently available with, can produce videos at 720p resolution and up to 8 seconds in length. The OpenAI solution can produce Full HD clips of up to 20 seconds.

Hallucinations still happen

Although the model is better in many aspects, Veo 2 still prone to hallucinations. As Google notes, AI can hallucinate details like extra fingers and unexpected objects. “Veo produces 2 [as alucinações] Gigante das Buscas argued that less frequency makes the results more realistic.

THE Security is also a focus for Google. The company is “intentionally measuring the expansion of Veo availability” to determine and understand the quality and reliability of the vehicle.

In this sense, Veo 2 integrates its own SynthID watermark into every result. These markers are invisible but not infallible.

Where does the data come from?

Another detail omitted from the announcement is the database used to train the model. It’s unknown where Google captures the videos to power the Veo 2, but YouTube is one possible source.

According to Eli Collins, DeepMind’s vice president of product, the AI ​​is trained using video pairs and visual descriptions.

To: Veo 2 functions will be integrated into Google Labs’ video creation tool VideoFX. The company said it is expanding the program to more users.

The use of the tool is free during this testing period. However, it is mandatory to sign up for the waiting list. You can do this on the official VideoFX website.