Tonight, OpenAI, the company behind ChatGPT, unveiled Sora, a text-to-video model designed to simulate the physical world in motion from a sentence, marking a significant advance in AI’s ability to blend in reality. Producing videos without an actor or set is no longer reserved for Pixar or Disney, anyone can get involved.
Update 09/12: OpenAI is launching Sora today for the general public. Information revealed by the famous YouTuber Marques Brownlee aka MKBHD who had already had access to the service for several weeks as a beta tester. Unfortunately, Europe is not affected at the moment.
Sora is a slap in the face
In fact, you just have to write a text for the video to be created by the AI, like DALL-E for generating images, but this time for videos. It is not the first artificial intelligence to offer it, Runway can for example animate an image with text, but it is clearly on a completely different level.
© OpenAI Sora
Sora can produce videos up to one minute long, maintaining high visual quality and strictly following user instructions. For any artist, filmmaker, or student who needs to create a video, this opens up many possibilities. The time saving is considerable, which allows you to test your ideas in the blink of an eye, before going deeper once the choice is made by returning to traditional methods. Entrepreneurs wanting, for example, to create a POC will find a strong ally there.
OpenAI specifies that Sora was made available to members of the Red Team (Red Team as opposed to Blue Team) to assess potential damage or risk, as well as to a select group of creative professionals to obtain a feedback on its usefulness in the professional context. OpenAI plans to refine Sora based on this feedback, to ensure that it effectively meets the needs of its users. The demonstration videos are absolutely stunning, not to mention surreal.
Sora can create complex scenes with stunning detail including multiple people, natural movements, and realistic environments. For example, Sora can create videos depicting an elegant woman walking down a neon-lit Tokyo street, giant woolly mammoths in a snowy meadow, a museum full of paintings, or a movie trailer showcasing the adventures of a space man.
It’s stunning 😳 https://t.co/GW1M6Jh8sY
— iPhoneSoft.fr (@iphonesoft_fr) February 15, 2024
OpenAI just announced Sora, their new AI video model!
If it wasn’t true before it is now, AI video is taking over in 2024.
9 insane examples from the new model:
— Proper 🧐 (@ProperPrompter) February 15, 2024
All of these UNBELIEVABLE videos were created using Sora, the new AI model from OpenAI
Watch each one and see how it makes you feel…
I don’t think it’s crazy for me to say this going to shift Hollywood, social apps and media forever
Video #1
Prompt: The camera directly faces… pic.twitter.com/NJfphtGbWb— GREG ISENBERG (@gregisenberg) February 15, 2024
The limits of the Sora AI
As a first release, Sora obviously presents several limitations. AI can get to grips with simulating the physics of complex scenes and understanding specific cause-and-effect scenarios. According to OpenAI, Sora can also confuse spatial details and have difficulty describing specific events in time.
On the security side, OpenAI is working with domain experts to test the model to detect misinformation, hateful content, and bias, as well as to develop tools to detect misleading content. The company also plans to integrate C2PA metadata in future releases to ensure content authenticity. Deepfakes are already difficult to spot, so with Sora…
Other useful features
Sora was created based on previous research of the DALL-E and GPT models. The feature can also transform an image and make it into a video, like Runway. Existing videos can also be extended or missing images completed. Pretty incredible.
In addition, know that Apple is also working on it with Keyframer, but we have not yet had a demonstration on it. WWDC 24 should be very interesting from this point of view…
