[ad_1]
\n<\/p>\n
This technical report focuses on (1) our method for turning visual data of all types into a unified representation that enables large-scale training of generative models, and (2) qualitative evaluation of Sora\u2019s capabilities and limitations. Model and implementation details are not included in this report.<\/p>\n
Much prior work has studied generative modeling of video data using a variety of methods, including recurrent networks,[^1]<\/span><\/sup><\/span>[^2]<\/span><\/sup><\/span>[^3]<\/span><\/sup><\/span> generative adversarial networks,[^4]<\/span><\/sup><\/span>[^5]<\/span><\/sup><\/span>[^6]<\/span><\/sup><\/span>[^7]<\/span><\/sup><\/span> autoregressive transformers,[^8]<\/span><\/sup><\/span>[^9]<\/span><\/sup><\/span> and diffusion models.[^10]<\/span><\/sup><\/span>[^11]<\/span><\/sup><\/span>[^12]<\/span><\/sup><\/span> These works often focus on a narrow category of visual data, on shorter videos, or on videos of a fixed size. Sora is a generalist model of visual data\u2014it can generate videos and images spanning diverse durations, aspect ratios and resolutions, up to a full minute of high definition video.
<\/p>\n<\/div>\n<\/div>\n