OpenAI announces "Sora," which generates up to a minute of video from text
On February 15, 2024, OpenAI announced “Sora,” its first model for generating video from text prompts. It released high-fidelity sample videos — two people walking through snowy Tokyo, an SUV driving down a mountain road, a fluffy monster next to a candle, and fictional footage styled after the gold rush era — demonstrating technology capable of generating videos up to a minute long.
Details
- Technical approach: A diffusion model that generates video as a whole rather than frame by frame. It applies a Transformer architecture over units called spacetime patches, designed to maintain consistency even when a subject moves out of frame
- Generation capabilities: Touted as able to generate video featuring complex scenes with multiple characters, specific types of motion, and precise details in both subjects and backgrounds
- Availability: Still in the research stage at the time of announcement, not yet integrated into any OpenAI product. OpenAI said it would begin by granting access to “red teamers” to assess harms and risks
- Technical report published: OpenAI simultaneously published a technical report titled “Video generation models as world simulators,” discussing the potential for video generation models to serve as simulators of the physical world
What happened next
About ten months after this research preview, Sora launched publicly as “Sora Turbo” for ChatGPT Plus and Pro users on December 9, 2024. It later evolved into the successor “Sora 2” and an accompanying social app in September 2025.