Google Deepmind has unveiled its next-generation AI tools, Veo 2 and Imagen 3, along with a playful new experiment, Whisk, aimed at remixing visuals in creative ways.

Veo 2 is the newest version of the company’s flagship video-generation tool, which according to Google will be capable of higher realism and an improved understanding of movement, physics, and cinematic techniques. In its press release, Google Deepmind states that Veo 2 can generate 4K videos and handle complex prompts — like specific camera lenses or cinematic shots.

“Veo 2 understands the unique language of cinematography,” the press release states, emphasizing its ability to deliver stylistic choices like “low-angle tracking shots” or “shallow depth of field.” Further, the company says that, unlike earlier models that often “hallucinate” strange visuals (like extra limbs), Veo 2 reduces these quirks, making results more natural.

Imagen 3, released in August, has gotten some improvements with richer textures, brighter colors, and diverse art styles — from photorealistic landscapes to anime. The upgrade will also stick more faithfully to user prompts.

The most intriguing addition might be Whisk, a tool that allows users to remix images with creative flair. By combining Google’s Imagen 3 and Gemini models, Whisk can turn rough image ideas into digital artworks like “plushies or enamel pins,” all while automatically generating descriptive captions.

Veo 2 is rolling out in Google’s VideoFX, Imagen 3 hits ImageFX in over 100 countries, and Whisk is live for U.S. users. You can sign up for the Google Labs waitlist now.