- Start
- Feb 15, 202490% CONFIDENCEfrom the source · possible Feb 15 – 16, 2024
Sora Previewed
OpenAI unveils Sora, a text-to-video model that makes photorealistic 1080p clips up to a minute long, but shows it only through demo videos and gives access to red-teamers and a few artists.
- On 15 February 2024 OpenAI previewed Sora, a text-to-video model it said could make videos up to a minute long from a prompt, and posted demos such as a snowy Tokyo street and a fluffy monster beside a candle[2][5]
- Ars Technica reports 1920x1080 output with a subject held consistent for up to 60 seconds, plus the ability to extend a clip or animate a still image, but no synchronized sound yet[3]
- It was not a product: only red-teamers and a few artists, designers and filmmakers had access, and an OpenAI spokesperson said the model would not be broadly available soon[4][5]
- OpenAI itself listed weaknesses in physics and cause and effect, and reviewers saw glitches in the demos, such as a Tokyo sidewalk that dead-ends and cars smaller than the pedestrians beside them[2][4][5]
- Wired and MIT Technology Review both caution that the clips were hand-picked and that render times were undisclosed[2][5]
- It became the first of five Sora releases that ran until OpenAI shut the app on 26 April 2026 and the API on 24 September 2026[6]
Translations completed using Gemini 3.8 Flash.
Buy onChatGPT