On Monday, Google announced the launch of Veo 2, the next generation of its video generation model. The new artificial intelligence (AI) model enhances the capabilities of its predecessor, which was released in early 2024. Notably, Veo 2 can generate videos in 4K resolution and extend video durations significantly. This introduction comes shortly after Google integrated the original Veo model into Vertex AI for enterprise solutions. At present, Veo 2 is accessible to a limited group of creators on the VideoFX platform through Google Labs.
Google Introduces Veo 2 with Enhanced Features
In a recent blog post, Google provided an overview of the advancements made with Veo 2. While the initial Veo model continues to serve users through Vertex AI and VideoFX, as well as in the background generator feature for YouTube Shorts, Veo 2 is reported to have achieved impressive results in internal testing, surpassing other video models like Sora.
Introducing Veo 2, our new, state-of-the-art video model (with better understanding of real-world physics & movement, up to 4K resolution). You can join the waitlist on VideoFX. Our new and improved Imagen 3 model also achieves SOTA results, and is coming today to 100+ countries… pic.twitter.com/FtwNKyicEc
— Sundar Pichai (@sundarpichai) December 16, 2024
Google claims that Veo 2 presents significant advancements in the detail and realism of its generated videos. The new model is said to possess an understanding of cinematographic language, enabling users to input specific genres, camera angles, cinematic effects, and movements for video production tailored precisely to their requests.
For instance, a sample prompt such as “a low-angle tracking shot with 18mm lens that glides through the middle of a scene” would lead to the generation of a wide-angle shot, showcasing Veo 2’s comprehension of camera techniques.
In addition to handling longer videos, with durations extending to “minutes,” Veo 2 boasts 4K resolution capabilities. The exact maximum duration of the generated videos has not been disclosed. Google has also stated that the model minimizes occurrences of hallucinations, significantly reducing the presence of extraneous details like extra fingers or distorted limbs in videos.
To address safety concerns, Google has incorporated the SynthID watermark into generated videos, a measure intended to distinguish AI-created content from real footage, aiding in the prevention of deepfake misuse. The company plans a phased rollout of Veo 2 to enhance quality and safety before full global availability.
Currently, users can access Veo 2 on VideoFX via Google Labs by joining a waitlist. The company also intends to extend availability to YouTube Shorts and other platforms in the coming year.
Imagen 3 Debuts for Global Access in ImageFX
In conjunction with the Veo 2 launch, Google has rolled out the Imagen 3 AI image generation model in ImageFX, now available in more than 100 countries. This latest version boasts several enhancements, including a broader range of art styles, improved accuracy, and superior composition.
Google asserts that Imagen 3 demonstrates increased adherence to prompts and can create images with more intricate details and textures. The model is particularly effective in styles such as photorealism, impressionism, abstraction, and anime.
