Google DeepMind has announced the updated version of its Veo AI video generator, Veo 2, which can produce videos up to 4K resolution. This marks an improvement over the original Veo, which was limited to 1080p. The new tool also boasts better “camera control” and physics in its scenes, allowing users to prompt specific shots and angles.
The announcement comes as Google DeepMind is competing with rival AI labs, including OpenAI’s Sora and Runway. However, Veo 2’s 4K resolution suggests that DeepMind may be pulling ahead in video generation capabilities. According to a study conducted by DeepMind, 59% of human raters preferred Veo 2 over Sora Turbo, with similar victories against other rival services.
The improved physics and “prompt adherence” features of Veo 2 are significant strides for the technology. However, some experts argue that issues like physics and object permanence can only be fully solved with world models that have spatial intelligence to understand and generate 3D environments.
To combat potential misuse, Google DeepMind has applied invisible SynthID watermarks to Veo 2 clips, making it more difficult to use them for nefarious purposes. In contrast, OpenAI’s Sora embeds a visible animation in its videos, which may not be as effective against malicious applications.
Veo 2 is now powering Google Labs’s VideoFX generation tool and the revised Imagen 3 model can be used in the ImageFX tool, with ImageFX available in over 100 countries. However, ImageFX remains unavailable in Germany due to technical reasons, according to a spokesperson for DeepMind.
Source: https://fortune.com/2024/12/16/google-deepmind-veo-2-ai-video-generator-4k-openai-sora