Artificial intelligence steps into a new world with OpenAI’s Sora

Text-to-video generator enters service... for some
Sora generates video based on text prompts

20 February 2024

The OpenAI website now features content created by text-to-video generator Sora, which can generate complex scenes with lots of detail in the subject and background, and with multiple characters. It can generate specific details because it has a good understanding of language. This also allows it to create vivid emotions and characters, as well as multiple shots within a video as if someone filmed it.

However, Sora is not the finished article. For example, it has trouble accurately mimicking the physics of a complex scene. It also does not yet properly understand certain cases of cause and effect. It also sometimes confuses details and has difficulty with precise descriptions of events taking place in time. This creates videos in which, for example, someone walks backwards, a chair floats or dogs multiply.

Currently, the program is available for security evaluation experts and a select number of creative professionals. OpenAI has also stressed that they cannot predict how people will use it.




OpenAI is implementing several security measures with Sora. For example, they will work with experts on misinformation, hate content and bias to test the model. In addition, Sora already uses existing security methods, but also creates new techniques to detect misleading content. This will allow the detection classifier to recognise videos created by Sora.

By these measures, the text classifier will check the prompts to see if it complies with the usage policy. The image classifiers will check the videos produced by Sora to see if it complies with the usage policy before showing it to the user.

News Wires

Read More:

Back to Top ↑