Sora

On the web at sora.chatgpt.com, Sora turns text prompts, still images or existing videos into AI-generated video clips up to a minute long.

Sora is an AI video generator and model built to follow prompts closely, and it can handle scenes with several characters while keeping visual style and characters consistent across shots.

You can also animate a still image or extend a clip you already have.

It runs on a transformer architecture, which is what lets it scale in quality.

The catch is physics.

Sora can get cause and effect wrong, so a cookie may show no bite mark after someone bites it.

It can mix up left and right, struggle with camera paths or events that unfold over time, and let objects morph oddly.

In crowded scenes, animals or people may appear from nowhere.

Expect to rerun and check results.