OpenAI announces Sora – AI that turns text into realistic video

Generative AI completely exploded in 2023, with both text- and image-generation taking giant leaps and causing ripples across entire industries. The adoption keeps spreading and there are no signs of a slow-down in 2024. In fact, it might actually accelerate. OpenAI just revealed their latest innovation – Sora. A generative AI model that takes text input and turns it into video and the results look very impressive.

Sora – Creating video from text

Despite fierce competition from tech giants like Google, OpenAI has maintained its lead in the generative AI space. ChatGPT is the fastest-growing platform of all time and has basically become synonymous with AI text generation. Now, ChatGPT is getting a younger sibling in the form of Sora. Sora is a video-generation model and according to OpenAI it can “create realistic and imaginative scenes from text instructions”.

Currently, the videos that are created can be up to a minute long and they are capable of creating some pretty complex scenes. The input can be anything from a short one-sentence prompt to more complex instructions where you dictate the location, video style, setting and go into tiny details.

A “prompt” is a text input provided by the user that instructs or guides the AI in generating a response or output.

Looking at how AI-generated pictures often have a hard time getting simple details right, like the number of fingers a person has, it’s impressive how realistic the videos Sora creates are. Sure, in some clips you can easily see that the videos are AI-generated and OpenAI themselves have said that Sora might struggle with getting the physics right in a complex scene. With other, however, it’s almost scary how realistic they are.

AI generated image of rocks hitting a rugged computer but failing to damage it
AI generated image of rocks hitting a rugged tablet.

No official release date

So far, there is no official release date for Sora and the public will have to wait. Right now only so-called “Red Teamers” can access the model. The Red Teaming Network is a network of expert volunteers that perform testing on OpenAI´s models to ensure they are safe and prevented from causing harmful content.

While it’s no fun waiting around to get to play with the shiniest tool on the internet, it’s for the best. There are already too many deep fakes circulating on the internet and while some are harmless, like the swagged-out pope, others the world would be better off without. It’s definitely important that the company acts responsibly and ensures that an AI video generative tool of this level is prevented from creating harmful or manipulative clips.

Sora already has competition

Sora is actually not the first video-generative AI on the market, the company Runway has already launched the second version of their AI video model. Google is working on its own video model called Lumiere. Meta, the parent company of Instagram and Facebook, has teased a new text-to-video model as well. Other tech giants in the AI business include Amazon and Elon Musk´s xAI, however, these two companies have not yet made any announcements regarding video generation. It remains to be seen whether Sora can become as dominant as ChatGPT.