Google Veo 2 and an updated Imagen 3: video up to 4K, one week after Sora
In one sentence On December 16, 2024 Google introduces Veo 2, a text-to-video model claiming output up to 4K and clips extendable to minutes, together with an improved Imagen 3 and the Whisk experiment. Veo 2 comes to VideoFX behind a waitlist.
On December 9, 2024 OpenAI opens Sora to the public. A week later Google responds with Veo 2, its new video generator, and an updated version of Imagen 3 for images.
According to Google, Veo 2 better understands real-world physics and human movement, gets fewer details wrong (such as extra fingers) and can produce video up to 4K resolution, extendable to minutes in length. In head-to-head comparisons with human raters, Google claims state-of-the-art results.
Access is gradual: first in VideoFX (Google Labs) behind a waitlist, then YouTube and Vertex AI, with YouTube Shorts expected in 2025.
Companies
Google, Google DeepMind
Tools
Veo 2, Imagen 3, Whisk, VideoFX, ImageFX
Tags
Sources