From ChatGPT to DALL-E and Now Sora; OpenAI Reveals A.I. That Turns Text Prompts to Realistic HD Videos

The U.S.-based artificial intelligence (AI) study group OpenAI just unveiled a new artificial intelligence text-to-video model; called Sora... in February 2024. It can make videos based on descriptive prompts and move parts of films forward or backward in time.

Sora creates complex scenes with several people, unique movements, and realistic subject and backdrop elements. The model knows both what the user requested in the prompt and how it exists in the physical world.

The model understands English well enough to interpret instructions and create intriguing characters with strong emotions. Sora may also maintain characters and visual style across several scenes in a created video.

The pros

One example of Sora's ability to figure out long questions (135-7 words) may be what makes it stand out. Thursday, OpenAI put out a sample movie that shows how flexible Sora is. There are many kinds of characters and scenes that it can make, such as people, animals, fluffy monsters, cityscapes, landscapes, zen gardens, and even New York City underground.

OpenAI says that Sora uses the recaptioning method from Dall-E 3, which creates "very detailed captions for the image training data."

"Sora is able to generate complex scenes with multiple characters, specific types of motion and accurate details of the subject and background," said the post. "The model understands not only what the user has asked for in the prompt, but also how those things exist in the physical world."

Some challenges

Open Artificial Intelligence admits that Sora isn't perfect. For example, she has trouble understanding cause and effect and getting the details of a complicated situation right.

Also, don't worry—Sora mixes up left and right too. This should help anyone who still has to make a L with their hands to figure out which is left.

No one knows when Sora will be available to everyone, but OpenAI said it needed to take "several important safety steps" first. This includes following OpenAI's current safety rules, which say you can't post violent or sexual content, hate speech, pictures of celebrities, or other people's intellectual property.

Key features

OpenAI noted that it will need to finish safety procedures before making Sora accessible to everyone, mentioning the AI model is being "adversarially" tested by professionals from areas like misinformation, hateful content, and bias.

The videos generated by Sora feature "highly detailed scenes, complex camera motion, as well as multiple characters with vibrant emotions," according to OpenAI, which blogged a string of Sora-generated footage which includes a stylish woman strolling down a the city of Tokyo street and a mock film trailer.

To keep an eye on OpenAI Sora when it gets launched to the public, visit their official website here