Sora 2

Sora 2 scales world simulation with 4K resolution, extended temporal consistency, and physics-compliant motion synthesis.

OpenAI’s Sora 2 advances the video generation landscape by doubling the sampling efficiency of the original transformer architecture. It renders complex scenes (up to 120 seconds) with precise object permanence and fluid camera trajectories. By integrating a refined spatial-temporal latent space, the model eliminates the warping artifacts common in earlier iterations. This version handles multi-character interactions and intricate fluid dynamics (like splashing water or pouring coffee) with high fidelity. It remains the industry benchmark for turning descriptive text into cinematic, high-resolution visual assets.

https://openai.com/sora