We all knew it was coming, but we didn’t expect it so soon.

OpenAI, famous for its ChatGPT and Dall-E free-to-use AI system, unveiled a new text-to-video service to select researchers, beta testers, developers and policymakers. “Sora” generates up to one-minute videos from text prompts. You type in a sentence or two as a prompt and it creates your video.

Check this out …

The Sora sample videos reflect realistic and other styles, producing digital video content for advertising, digital signage, social media, corporate websites — for everything and anywhere video appears.

Hollywood is safe, as there are processing power limitations — it can only create about a minute of content. But when OpenAI turns even this small tap on, it will flood social media and create a new digital world. Video by anyone. And everyone.

If you are old enough, you’ll remember how “desktop publishing” changed the world. What Sora has now wrought is a 10x bigger disruption factor.

With realistic scenes that include complex interactions, multiple characters and detailed backgrounds, Sora distinguishes itself with emotionally expressive characters with consistent style and character presence across multiple video shots.

In short, it seems very real.

Sora serves as a foundation for models that can understand and simulate the real world, a capability OpenAI believes this will be an important milestone for achieving Artificial General Intelligence. Sora can also extend existing videos with new frames or take an existing still image and make a video from it by animating the content with accuracy and attention to small detail.

This is controversial wizardry that scares the hell out of some content owners while other content creators cheer it on. OpenAI knows this. The company will take several critical safety steps before making Sora available by subscription (similar to ChaptGPT, Dall-E and its AI product family).

The company has turned to “red teamers ” ( domain experts in areas like misinformation, hateful content and bias )  who will run “adversarial” tests. OpenAI are also busy building tools to help detect misleading content, tools like a detection classifier that can tell when a video was truly generated by Sora. According to OpenAI, they will include C2PA metadata if they deploy the model in an OpenAI product.

I will admit, however, that Sora 1.0 is not perfect. Despite its innovative approach, it still faces challenges with complex physics, accurate cause/effect representation and maintaining precise spatial/temporal details.

Yet Sora is already a giant leap forward in digital video content, a leap that might cross Moore’s chasm before proper harnesses can prevent its misuse.

Want to get technical about Sora? Here’s how OpenAI achieves it technically:
https://openai.com/research/video-generation-models-as-world-simulators

More about Sora
https://openai.com/sora