Google updates its Veo and Imagen AI models for more accurate content
Google has announced a new version of its generative AI model for video, Veo 2, which the company says now better understands real-world physics and human movements.
Here's What We Know
Veo 2 allows users to reference specific film genres, cinematic effects, and lenses when creating videos. The model also reduces artefacts such as extra fingers and improves image quality. Below is a video created entirely by the Veo 2 model.
Google has also improved its text-to-image model, Imagen 3, which now generates brighter and better composed images, and follows cues more accurately.
An image generated by the Imagen 3 model. Illustration: Google
Google has also added a new tool called Whisk, which combines the capabilities of Imagen 3 and the Gemini visual understanding model to create unique images by combining multiple ideas or objects into a single illustration.
The models include an invisible SynthID watermark to reduce the likelihood of misinformation. Veo 2 will be gradually available to Google Labs users in the US and is now limited for testers for now to create videos of up to eight seconds in 720p. Enhancements for Imagen 3 are already available to Google Labs users in more than 100 countries through ImageFX.
Source: Google
