Google DeepMind has announced Veo 2, its latest AI video generation model, positioning it as a direct competitor to OpenAI’s Sora. The new model is currently available through Google Labs’ VideoFX platform on a waitlist basis, with users required to apply through a Google Form for access.
According to Google, Veo 2 can generate videos up to two minutes long in resolutions reaching 4K (4096 x 2160 pixels), though the current implementation in VideoFX limits outputs to 720p resolution and eight-second clips. The company claims the model offers improved understanding of real-world physics, human movement, and cinematographic controls, including the ability to specify genres, lenses, and cinematic effects.
The new model incorporates several technical improvements over its predecessor, particularly in areas such as motion rendering, fluid dynamics, and light properties. Google states that Veo 2 produces more realistic textures and sharper imagery, especially in scenes with significant movement. The company reports that in internal tests, human evaluators preferred Veo 2’s outputs over those of competing models, including Sora, in terms of overall preference and prompt adherence.
To address potential misuse, Google has implemented its SynthID watermarking technology to identify AI-generated content. The company acknowledges that while Veo 2 represents an advancement in video generation capability, it still faces challenges with hallucinations, such as incorrect anatomical features, and maintaining consistency in complex scenes.
The release comes amid increasing competition in the AI video generation space, with various companies including RunwayML, Pika Labs, and Luma AI releasing their own solutions. While some filmmakers and creators have begun embracing these technologies, there remains skepticism in the creative community regarding the consistency and reliability of AI-generated video content.
Sources: Google DeepMind, VentureBeat, TechCrunch