OpenAI Sora: New text-to-video AI model delivers incredible results

The US software company OpenAI has presented a new AI model. Sora converts text prompts into complex video scenes that are up to one minute long. These can contain different characters and deliver both realistic and imaginative results that are almost impossible to put into words. And this despite the fact that the videos themselves are based on words. The second scene in the following video is based on the following text prompt: "A litter of Golden Retriever puppies are playing in the snow. Their heads are sticking out of the snow, covered in it."

If you want to know which text input led to the respective result, you can find the individual videos and prompts on the OpenAI website.

Sora not only has a deep understanding of language to interpret the input, but also knowledge of how things behave in the physical world. Nevertheless, the videos are still far from perfect. If you look closely, you will discover the odd mistake.

OpenAI knows this too. The company points out that the physics have weaknesses in complex scenes. Spatial details can lead to confusion - as can temporal sequences such as a tracking shot. OpenAI also points out that it is possible that Sora does not understand certain cases of cause and effect: "For example, a person might bite into a biscuit, but the biscuit might not show a bite mark afterwards."

Technically speaking, Sora is a diffusion model that is able to create entire videos at once or extend an existing one. It is also possible to use a still image as a template instead of text input. So far, Sora is only available to a selected group of testers. It is not yet clear when the AI model will be released to the public and at what price.