Google has unveiled a preview of Veo 2, a significant update to its generative AI video tool, just after beginning its enterprise rollout of the original Veo. This new version is designed to “understand the language of cinematography,” allowing users to prompt the model using specific film genres, cinematic effects, or lens types.
Read: Bandwidth Blog & Smile 90.4FM Tech Tuesday: Moon base missions!
Google also claims Veo 2 demonstrates improved understanding of real-world physics and human movement, areas where generative models often struggle. While the provided examples are insufficient for definitive proof, the claim of improved motion rendering, particularly for complex actions like a gymnast’s routine, is noteworthy. Google also states that Veo 2 will produce artifacts like extra fingers “less frequently,” another common issue with video generation models.
In addition to Veo 2, Google is rolling out enhancements to its text-to-image model, Imagen 3. The company says the latest version produces brighter, better-composed images and renders diverse art styles with greater accuracy. It also demonstrates improved prompt adherence, an issue previously highlighted when Imagen 3 was released to Google Cloud customers.
Veo 2 will be gradually released to Google Labs users in the US, initially limited to generating eight-second clips at 720p. This contrasts with OpenAI’s Sora, which can generate up to 20 seconds of 1080p video, though access is tied to a $200 per month ChatGPT Pro subscription. The Imagen 3 enhancements are available to Google Labs users in over 100 countries through ImageFX.
