Intermediate · Generative AI

Sora

Visual diagram · (in preparation) · Math · (in preparation) · Worked example · 3 difficulty levels.

TL;DR. OpenAI's text-to-video diffusion-transformer model that generates minute-long high-resolution clips with strong temporal coherence and 3D consistency.

Technical Definition

OpenAI's text-to-video diffusion-transformer model that generates minute-long high-resolution clips with strong temporal coherence and 3D consistency.

How it works

Sora treats video as sequences of spacetime patches and processes them with a Diffusion Transformer (DiT). Training on internet-scale video with recaptioned descriptions yields emergent simulation-like behavior: object permanence, plausible physics, and camera motion. Released publicly in late 2024, Sora is OpenAI's flagship demonstration that scaling generative video follows scaling-law trends seen in LLMs.

Related Concepts

• Diffusion Model — A generative model that learns to create data by reversing a gradual noising process, producing high-quality images and audio.
• Text-to-Image — A generative task that produces images from natural-language descriptions.
• Diffusion Transformer (DiT) — A diffusion model that replaces the U-Net denoiser with a Transformer operating on image or video patches, scaling cleanly with compute.