Google DeepMind, the search giant’s artificial intelligence area, seems to want to hit the table every time OpenAI presents a new product. And now, It’s the video’s turn with its new model: Veo 2.
From its predecessor, Veo, to the new version, DeepMind has made an impressive leap. Not only does Veo 2 allow for longer clips (over two minutes) and higher resolutions (up to 4K), but it has also improved your understanding of concepts such as fluid dynamics and the properties of light. Come on, the examples you have shown are really impressive.
Veo 2 has superior ability to model complex movementssuch as the flow of a liquid or the movement of a camera that follows an object from multiple angles.
Today, we’re announcing Veo 2: our state-of-the-art video generation model which produces realistic, high-quality clips from text or image prompts. ?
We’re also releasing an improved version of our text-to-image model, Image 3 – available to use in ImageFX through… pic.twitter.com/h6ejHaMUM4
— Google DeepMind (@GoogleDeepMind) December 16, 2024
From imitating cinematographic styles to touching the ‘uncanny valley’
One of the new features is the possibility of applying cinematic styles and refined details: AI can replicate styles like that of Pixar animation or effects such as shadows and reflections on a stage.
However, despite these improvements, the tool has some problems. For example, some generated videos show “uncanny valley”, a phenomenon in which faces or movements do not seem entirely realwhich can be disturbing to viewers.
Google plans to integrate Veo 2 into VideoFX, its experimental video creation tool, and Vertex AIyour platform for developers. With these integrations, the aim is to democratize access to technology and allow developers and content creators to experiment with this innovative AI.
DeepMind has already worked with artists such as Donald Glover and The Weeknd to better understand the needs of the creative industry and fine-tune its models.
Google believes that In the future, its AI can help generate trailers and visual effects to the creation of educational or promotional content, the possibilities are vast.
Google and its connection with YouTube raise alarm bells
However, Google also already has criticisms related to using data to train your models. Although it claims to use “high-quality video pairings and descriptions,” the lack of transparency about the sources of this data, possibly including YouTube content, has raised copyright concerns.
AI video generation raises complex ethical and legal questions. I see 2, although impressive, is not without risks such as deepfakes and misinformation: Despite the use of technologies such as SynthID to insert invisible watermarks into generated clips, they are not foolproof.. This could facilitate the creation of deepfakes or maliciously manipulated content.
DeepMind claims to be committed to ethical collaboration with the creative community, although this commitment will be put to the test as the tool gains popularity.
#Googles #bet #generate #videos #artificial #intelligence #examples #scary
