- Google has released its video generation model called Veo 2. It can generate minutes-long videos at 4K resolution.
- The search giant has also unveiled Imagen 3 which is an image generation model.
- Besides that, Google has introduced an experimental "Whisk" AI tool that remixes subjects, scenes, and styles to create unique images.
Google has finally released its AI video generation model called Veo 2 to rival OpenAI’s Sora model. Unlike Sora which can generate 20-second clips up to 1080p resolution, Google’s Veo 2 model can produce minutes-long videos at stunning 4K resolution. Importantly, Google says Veo 2 understands “real-world physics” and “the nuances of human movement and expression” for improved realism.
In our Sora hands-on testing, we noted that OpenAI’s video generation model lacks a good understanding of physics. But Veo 2 promises better output as it hallucinates less. For example, with Veo 2, you are unlikely to get extra fingers in humans or unexpected objects in the final output.
Apart from that, you can generate videos in different styles by defining the lens, genre, cinematic effects, close-up shots, low-angle tracking shots, and more.
Veo 2 is being slowly rolled out via VideoFX, YouTube, and Vertex AI. To access Veo 2, you need to sign up for the waitlist. Google says Veo 2 is coming to YouTube Shorts next year. And all Veo 2-generated videos will include an invisible SynthID watermark to identify AI-generated videos.
Besides video generation, Google has also announced Imagen 3, its AI image generation model. Imagen 3 can now generate “brighter, better composed images.” It’s also more accurate and can produce a wide range of art styles including abstract, anime, photorealism, and impressionism. It’s widely rolling out on ImageFX in over 100 countries.
Finally, Google has introduced a new experimental tool called “Whisk” which lets you add a subject, scene, and style to create a unique image by remixing them. For example, you can upload a photo as your subject, then you can add a scene or enter a prompt to describe the scene. Next, you can define the style and add a prompt to produce images you have in mind.
It uses Imagen 3 and Gemini’s visual understanding to blend all the inputs and generate a new image. You can access Whisk on Google Labs.
