OpenAI, the company behind ChatGPT, has developed an artificial intelligence model called Sora that creates videos of up to 60 seconds from detailed text instructions. The clips are of high quality, which shows that technology has advanced a lot in the last year. “Sister” is not yet available to the general public as negative effects are being considered.

This AI model will initially be available to a small number of experts to test how the technology can be used for nefarious purposes. Many negative scenarios will be analyzed to see how serious situations can be avoided.

• You can watch some demo videos here

OpenAI has released some short videos showing what this model, called Sora, is capable of. You can give it many commands, from “show me woolly mammoths running in the snow” to “make a video of wild animals competing in bike races.”

OpenAI also has a program called Dall-E that creates images based on user instructions. DALL-E was announced in January 2021, DALL-E 2 in April 2022, and DALL-E in October 2023.

The clips on Sora are not perfect, sometimes the movements are strange or “plugged in”, but the technology has a lot of potential, it has advanced a lot in the last year, and clips can last a maximum of 60 seconds, while the first clips created with AI technologies can be much longer.

The company realizes that the technology has a huge potential to be used to deceive people, especially since the phenomenon of “deepfake” has gained momentum and is becoming increasingly risky in the context of elections in many countries. OpenAI has a Security section in the Sora related notice.

OpenAI says it is also working on technologies to analyze metadata to quickly detect when a video was created using AI technologies.

Also, text instructions asking for violent or sexually explicit videos or requests with celebrity names are not accepted.

There is a lot of discussion about tagging AI-generated clips with special marks (watermarks) to make them easy to identify.

Sources: Washington Post, Wall Street Journal