Readme
Veo 2
Veo 2 is a video generation model designed to create videos with realistic motion and high-quality output (up to 4K resolution). The model can be accessed via VideoFX and Vertex AI.
Key Features
Quality and Control
- Follows both simple and complex instructions
- Simulates real-world physics
- Supports a wide range of visual styles
Technical Capabilities
- Enhanced realism and detail with reduced artifacts
- Advanced motion representation with accurate physics simulation
- Precise interpretation of camera instructions (shot styles, angles, movements)
Benchmarks
Veo 2 has been evaluated against other video generation models through human ratings on the MovieGenBench dataset (1003 prompts, released by Meta). Testing parameters: - 720p resolution for all comparisons - Veo samples: 8 seconds duration - VideoGen samples: 10 seconds duration - Other models: 5 seconds duration
Known Limitations
- Challenges with maintaining complete consistency throughout complex scenes
- Difficulty with complex motion
- Some limitations in creating highly realistic or intricate videos
Development Credits
Veo 2 was developed by a team at Google DeepMind. Key research and engineering contributions from:
Agrim Gupta, Ali Razavi, Ankush Gupta, Dumitru Erhan, Eric Lau, Frank Belletti, Gabe Barth-Maron, Hakan Erdogan, Hakim Sidahmed, Henna Nandwani, Hernan Moraldo, Hyunjik Kim, Jeff Donahue, José Lezama, Kory Mathewson, Kurtis David, Marc van Zee, Medhini Narasimhan, Miaosen Wang, Mohammad Babaeizadeh, Nelly Papalampidi, Nick Pezzotti, Nilpa Jha, Parker Barnes, Pieter-Jan Kindermans, Rachel Hornung, Ruben Villegas, Ryan Poplin, Salah Zaiem, Sander Dieleman, Sayna Ebrahimi, Scott Wisdom, Serena Zhang, Shlomi Fruchter, Weizhe Hua, Xinchen Yan, Yuqing Du, Yutian Chen
Additional acknowledgments to numerous partners across Google DeepMind and Google.
Related Technologies
- Imagen (text-to-image model)
- Gemini (AI model)
- SynthID (watermarking and identification for AI-generated images)