🖼️ Google's Veo 2: a competitor to Sora
Google DeepMind has revealed Veo 2, its state-of-the-art video generation model.
Veo2 generates videos up to 8 seconds long in resolutions of up to 720p. DeepMind claims that the model can create 4K videos up to 2 minutes in duration.
Main features:
🎥 In Google tests, Veo2 beats Sora, Kling, and Minimax in more than half of the tasks;
🎥 Minimum of hallucinations and artifacts; real-world physics in videos;
🎥 Advanced camera control: complex motions, a wide range of styles, angles, and combinations of all of these.
Despite significant progress, the developers admit that even Veo 2 is far from always coping with complex physics and scenes.
Some Veo 2 clips — on the video ⤴️
The model is available in experimental mode via VideoFX. You can sign up for the waitlist here.
📸 DeepMind has also updated its Imagen 3 image generator. The first version produced results on a level between Flux 1.1 pro and Ideogram V2. The updated Imagen 3 beats even the Text-to-Image Arena leader Recraft V3 in Google's benchmarks.
The latest version of Imagen 3 is already running in Gemini.
More on the topic:
