OpenAI 官方博客T1·Models
92 pts
OpenAI officially unveiled Sora, a text-to-video generation model, proposing the thesis that 'video generation models are world simulators.' Built on a diffusion transformer architecture and trained on large-scale video data, Sora generates high-fidelity, coherent videos up to one minute long that adhere to physical laws. This breakthrough marks a critical leap from static image generation to dynamic spatiotemporal simulation.