A shot from a video of a woman walking down a Tokyo street generated by Sora
OpenAI
Sora is an artificial intelligence video generation tool. It’s like Dall-E or Midjourney, but for motion rather than still images.
OpenAI first previewed the software in February 2024, and has finally released it to the public. Until now only a select few have had access to Sora.
The bad news: it’s not available in the UK yet.
“Users can access Sora everywhere ChatGPT is available, with the exception of the United Kingdom, Switzerland and the European Economic Area. We are working to expand access further in the coming months,” reads the OpenAI blog.
Those outside the UK who subscribe to ChatGPT Plus or Pro can use Sora, to generate clips up to 20 seconds long.
This isn’t quite the version of Sora we saw in February either, or the model that has powered the many examples of Sora-generated video you may have seen on social networks earlier in the year. Sora Turbo is a new model, one “significantly faster” than the original Sora.
It means Sora Turbo takes less processor time to generate a video, making it more viable for a wide release. Does this also mean it’s less impressive than the original? We’ll have to see what the AI enthusiasts are able to make using it.
What is Sora?
Sora is a new AI diffusion model from OpenAI, the company that’s become a poster child for the modern field of artificial intelligence.
These types of software can create new content that closely resembles the data they were trained on. In the case of Sora, the AI can produce hyper-realistic videos up to a minute in length based on the instructions you type. It can also extend videos to make them longer.
AI diffusion models are another form of generative AI. They work by “destroying” an image by turning it into indistinguishable pixels, and then reversing the process to create a coherent image. One way to look at it is the AI is crumpling and smoothing out a piece of paper with a doodle on it.
Once trained, a diffusion model can be guided by users to craft highly detailed visuals from clumps of data.
What can Sora do?
In OpenAI’s words: “Sora is able to generate complex scenes with multiple characters, specific types of motion, and accurate details of the subject and background.
“The model understands not only what the user has asked for in the prompt, but also how those things exist in the physical world.”
The results are jaw-dropping, to say the least. To show what Sora is capable of, OpenAI shared a series of realistic demo videos including scenes of humans and animals in motion, a tracking shot of a vehicle on a dirt road, period footage of California during the gold rush, CGI red pandas in a petri dish, and an interior of an art gallery.
As with other AI images, look closely and you may spot some errors, like a cat with an extra paw or a floating chair that makes you feel like you’re watching a glitched-out video game.
OpenAI says Sora is trained on data with variable durations, resolutions, and aspect ratios, which it claims can result in more accurate representations of reality.
Meanwhile, Al boffs reckon it may have been hooked up to Unreal Engine 5, a video game engine designed to build interactive 3D experiences. The giant toolbox provides the core elements, from physics to lifelike lighting and sound design, for developers to create more realistic-looking games.
More broadly, OpenAI believes that video generation models like Sora offer “a path towards building general purpose simulators of the physical world”.
We’re waiting for a call from Morpheus any minute now.
Is Sora out now?
Sora is available to subscribers of ChatGPT Plus and ChatGPT Pro, but not those in the UK or Europe yet.
Plus subscribers are able to create up to 50 videos at 480p resolution, the resolution of the old DVD format, per month. Or “fewer” at 720p resolution. These clips can be up to 20 seconds long.
ChatGPT Pro subscribers get “10x more usage,” higher resolution options and longer clip runtimes, but OpenAI says the full potential of Sora will be fleshed out in 2025 with “tailored pricing for different types of users.”
It’s not hard to picture wannabe AI filmmakers planning on going hard on Sora, before polluting the internet with their creations.
As with its other AI tools, OpenAI says it will ban clips that fall foul of its content guidelines, including videos that contain extreme violence, sexual content, hateful imagery, celebrity likeness, or the intellectual property of others.
OpenAI also warns this is far from a perfect product, should you be planning on creating a Michael Bay style epic with it.
“The version of Sora we are deploying has many limitations. It often generates unrealistic physics and struggles with complex actions over long durations,” says OpenAI.