OpenAI, the artificial intelligence research company behind large language models like GPT-3 and image generators like DALL-E 2, has set its sights on an even more ambitious frontier – generating photorealistic videos entirely from text. The company recently unveiled Sora, their new text-to-video AI model that they claim can create imaginative and realistic scenes based solely on written prompts.

With Sora, users can input text descriptions and the model will generate video clips of up to one minute in length, bringing their words to life in stunning visual detail. According to OpenAI, Sora can interpret complex scenes involving multiple characters, specific types of motion, and intricate details of the subject matter and backgrounds.

Sora is a tool that can make detailed pictures with many people in them, doing different things, and including lots of correct details about both the people and the place where the action is happening. OpenAI, the group that made Sora, says that this tool is also really good at understanding how things are supposed to look in real life. It can also make pictures of characters that look real and show strong feelings.

But text-to-video is not Sora’s only trick. The model can also generate new videos from an existing still image, fill in missing frames for videos, or extend existing clips into longer narratives. This flexibility gives Sora broad potential applications in fields like filmmaking, animation, and game development.

While groundbreaking, Sora is not without its limitations. OpenAI concedes the model “may struggle with accurately simulating the physics of complex scenes.” Some of the demo videos reveal small imperfections, like suspiciously moving floors in a museum scene. But overall, the results are undeniably impressive and hint at a future where text could be routinely transformed into moving images.

Just a couple of years ago, text-to-image generators like DALL-E and Midjourney garnered significant attention for their novel ability to create coherent visuals from written prompts. However, recent advances in AI systems have accelerated the development of text-to-video models at a remarkable pace.

Other companies like Runway and Pika have showcased their own impressive text-to-video models, and Google’s Lumiere system figures to be a primary competitor to Sora as well. Similar to OpenAI’s offering, Lumiere provides users with tools to generate videos from text and can also create videos from still images.

OpenAI is proceeding cautiously with the release of Sora. Currently, access is restricted to a small group of “red teamers” who are tasked with assessing the model for potential harms and risks before it sees wider release. The company is also soliciting feedback from select visual artists, designers, and filmmakers as they evaluate the system.

The existential risks posed by sophisticated generative AI models remain a looming concern. Just this month, OpenAI announced it would start watermarking images created by DALL-E in an effort to curb misinformation, though they conceded these markers could easily be removed. As lifelike synthetic videos become more ubiquitous, society will have to grapple with the consequences of this powerful technology.

But for now, OpenAI wants to put Sora’s awe-inspiring visual creativity on display. With this new model, the frontiers of AI language models are rapidly evolving, pushing into territory that was unimaginable just a few years ago. The boundaries of what these systems can create seem to expand with each new breakthrough. As Sora demonstrates, AI may be approaching a “make-anything” capability – where text can conjure not just images, but entire worlds in motion.