OpenAI, the maker of ChatGPT, announced its latest artificial intelligence engine for creating videos from text prompts.
OpenAI already has Dall-E for generating still images, but now we've moved to video, completing a suite of AI tools for any creative purpose.
Sora AI is not yet readily available to the general public. Still, in February 2024, it became available to a select few people whose job is to test the security and stability of AI products, also known as “red teamers.”
However, OpenAI founder Sam Altman has demonstrated Sora's capabilities by responding to people's instant requests with a final product produced by a video generator, which has already generated a lot of buzz on social media. I'm here.
Early signs indicate that this will be as impressive as ChatGPT and Dall-E, and definitely represents a new era of text that will inspire filmmaking.
What is Sora AI?
Just as we're familiar with text and still image generation AI models like ChatGPT, Dall-E, and Google Gemini, Sora AI does the same for video.
The AI works like any other generative AI model. It is constantly learning from what it sees and consumes, and is trained to give you the most accurate and detailed response to any prompt.
Sora AI is no exception. For example, if you enter a text prompt such as “Blue boat sailing on the ocean in the sun,” it will return that exact video. You can be as specific or vague as you like, the more detail you give the AI model, the better the results will be.
Demonstration of Sora AI in X by Sam Altman In February, there will be enough detail in the text prompt to best understand how it works and how accurate it is.
For example, the video below is the result of the prompt “Homemade gnocchi cooking instruction session hosted by a social media influencer's grandma, set in a rustic Tuscan countryside kitchen with cinematic lighting.”
https://t.co/rmk9zI0oqO pic.twitter.com/WanFKOzdIw
— Sam Altman (@sama) February 15, 2024
How does Sora AI work?
The technology behind Sora AI is the same technology that makes search on the internet possible. The more examples we see of AI, the more it will be able to spit out the same thing in other images. Eventually, when an AI understands one thing well enough, it can generate its own version on demand.
Of course, this is a very simplistic way of explaining how generative AI works, but OpenAI has previously provided a more detailed explanation of how its AI models work . Sora AI is trained on publicly available and licensed data You can see what your video will look like on a realistic level. It's trained to know what it's looking at, and learns how to use that information to generate its own version.
When you ask Sora AI to create a dog video, Sora AI generates results based on all the dog videos it has ever seen. Use visual patches and building blocks to help you understand which elements of your video go where from frame to frame. The more you watch and learn, the better and more accurate you become.
Sora's technology is built on a diffusion model, where the AI starts with a confused response before improving its output through a series of feedback loops. We also use a number of data analysis techniques to process vast levels of data before using Transformer technology to understand which important parts of the video to keep and unimportant details to omit.
What can Sora AI do?
To date, Sora AI can generate up to 1 minute of HD video from a text prompt. It can produce “real world”, cartoon, and CGI style videos, but cannot currently include audio.
Sora AI can also generate videos from still images, fill in missing frames in existing videos, and stitch multiple videos together. There is also the ability to generate infinite loops.
There are also examples of creating simulations of video games such as Minecraft.
There are reportedly plans to add audio and editing tools to Sora AI, the latter of which will allow creators to manually fix errors in the AI's videos. Perhaps the AI will learn from the manual corrections made, but that's just a guess and nothing more at this point.
However, there are some expected limitations. OpenAI recognizes these limitations, such as people disappearing, turning into other objects, and moving in ways that would be impossible in the real world. OpenAI has confirmed that it is already working on fixes for these issues.
Once these issues are resolved and Sora AI becomes more sophisticated and accurate, it can be used to create authentic fantasy worlds and movies, or to map real places in the world without the need for artificial intelligence. We cannot completely rule out the possibility of being able to explore. Visit them physically.
How do I access Sora AI?
Sora AI is not yet available to the public without an invitation. Individual authors and testers are encouraged to test with AI models to ensure that OpenAI is working based on their feedback and is ready for public release.
There are also very important aspects: security and ethics. Generative AI models have been exploited by criminals and pranksters in the past, such as with Taylor Swift's sexually explicit deepfakes, so OpenAI is ensuring that Sora AI can only be used for benevolent and creative purposes. I put a lot of effort into doing that.
This means you can no longer use Sora AI to generate videos that contain extreme violence, sexual content, hateful images, or likenesses of celebrities. OpenAI also includes metadata in Sora AI that indicates the video was generated by AI.
Mira Murati, OpenAI's chief technology officer, suggested in an interview with the Wall Street Journal that Sora AI will follow the same expedited policy as Dall-E. This means refusing to make videos of celebrities.
It's still unclear when exactly Sora AI will be available to the public, but Murati suggested it could be near the end of 2024. On the other hand, there is some speculation that when Sora AI is released, it will be released as a web app. Eventually, it will include additional features similar to ChatGPT, such as custom bots. It will face stiff competition. Microsoft recently released VASA-1, a new AI-powered tool that converts still images into clips.
Featured image: OpenAI // Sora AI
