OpenAI just introduces a new video-making tool called Sora. The company says Sora can make realistic videos based on written instructions. With Sora, users can create realistic videos up to one minute long just by writing prompts.

According to OpenAI’s first blog post about Sora, it can make complicated scenes with many characters, different movements, and detailed backgrounds. The company also says that Sora can understand how things are in the real world and create interesting characters with lively emotions.

Advancements in Video Generation Technology

The Model can make a video from just a picture and add new frames to make a video longer. In OpenAI’s blog post, they show videos made by Sora, like one showing California during the gold rush from the air, and another that seems like it’s taken from inside a train in Tokyo.

Some of these videos have hints that they were made by AI, like a floor that moves strangely in a museum video. OpenAI says Sora might not always get the physics right in a complicated scene, but overall, the results are impressive.

A few years back, models like Midjourney were leading the way in turning words into images. But now, video technology is advancing quickly. Companies like Runway and Pika are showing off their own impressive text-to-video models. Google’s Lumiere is also expected to compete with OpenAI in this area. Like Sora, Lumiere allows users to turn text into videos and even create videos from a single image.

Right now, Sora is only accessible to “red teamers” who are evaluating the model for possible risks and issues. OpenAI is also giving access to some visual artists, designers, and filmmakers to gather feedback. They mention that the current version of the model might not accurately represent the physics of complicated scenes and might not understand certain cause-and-effect situations correctly.

Earlier this month, OpenAI said it’s putting watermarks on its DALL-E 3 tool, which turns text into images. However, they mention that these watermarks can be taken off easily. Just like their other AI tools, OpenAI will have to deal with the problem of people mistaking fake, realistic-looking videos made by AI for real ones.

Conclusion

OpenAI’s new tool Sora makes realistic videos from text, but it’s still learning. Competitors like Google’s Lumiere show the field is changing fast. Sora’s only for testing now, and there are still some issues to iron out, like getting the physics right.

The post OpenAI Introduces ‘Sora’ AI Model That Turns Text Into Video first appeared on Business d'Or.