Sora: OpenAI unveils realistic video generation

2024-02-15 · Publisher : OpenAI · Domain : Artificial Intelligence

OpenAI unveils Sora, a model that turns a plain text prompt into realistic videos up to one minute long. Access is initially limited to red teamers and a handful of creative professionals.

On February 15, 2024, OpenAI unveiled Sora, a text-to-video generation model. The clips released that day, including a walk down a Tokyo street, stood out for their realism and length, far beyond what publicly available tools offered at the time. ### What happened - OpenAI announced a model able to generate videos up to one minute long, with detailed scenes, complex camera motion and multiple characters. - The model was not opened to the public: it was given to red teamers (testers probing for risks such as misinformation and bias) and to artists, designers and filmmakers for feedback. - OpenAI also published a technical report describing video generation models as world simulators, while acknowledging flaws: sometimes inconsistent physics, left and right mixed up, poorly rendered cause and effect. - The company announced classifiers able to detect a Sora-generated video and the addition of C2PA metadata. ### Key figures | Item | Value | |---|---| | Announcement | February 15, 2024 | | Announced maximum length | 1 minute | | Public release (Sora Turbo) | December 9, 2024 | | Length and resolution at public launch | 10 to 20 seconds, 480p to 1080p | | Availability at public launch | ChatGPT Plus and Pro subscribers, excluding the EU, UK and Switzerland | ### Why it is a turning point Sora brought video into mainstream generative AI, as DALL·E had done for images. The announcement accelerated the race among competitors (Google, Runway, Kuaishou and others) and put deepfakes and content provenance at the center of the debate, months ahead of major elections in 2024. ### Caveats The February announcement relied on demos selected by the vendor, with no public access or independent evaluation. The product released in December was a faster version (Sora Turbo), shorter and lower-resolution than the initial clips, and its European launch was delayed. ### Key takeaways Sora marks the moment realistic video generation became credible. For businesses, it opens up uses (visual prototyping, communication) and calls for provenance and labeling rules for generated content.

Tags : IA, Vidéo, IA générative, OpenAI, Sora