OpenAI forays into generative videos through text

GenVideos: ChatGPT maker OpenAI has launched Sora, a generative video model that converts short text descriptions to minute-long videos.

Taking to X, CEO Sam Altman shared the link to the video generation model, which is currently available to select creators and cybersecurity professionals.

OpenAI has not unveiled a date for making the service available to all users.

Pioneering a major advancement in text-to-video generation, the foray is OpenAI’s next step after it revolutionised codes to mails through short commands.

Notably, text-to-video services are already made available by Synthesia, DeepMind and Canva among others. Facebook-parent Meta has also forayed into the technology.   

“Sora can 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,” a description on the dedicated microsite reads.

Reflecting Sora’s capabilities, a 64-word prompt resulted in the creation of a high-definition, detailed one-minute video that looked remarkably real.

The video depicts a woman walking down the streets of Tokyo. The prompts were detailed, with what the woman was supposed to wear, how she is supposed to walk, along with a feel of the environment.

Sora was developed by integrating a diffusion model, similar to OpenAI's text-to-image model DALL-E 3, with a transformer neural network which allows Sora to process video data by dividing it into spatial and temporal chunks, enabling it to understand and generate coherent sequences with diverse styles and resolutions.

Commenting on the X update, renowned YouTuber Jimmy Donaldson who goes by the alias “Mr. Beast” said “Sam please don't make me homeless.”

Altman also reposted CRED CEO Kunal Shah's prompt, which showed a bicycle race on the ocean with different animals as athletes, taken through a drone camera view. 

Altman, who is known to be a cautious innovator pioneering newer technologies, has implemented safety measures, including filters to block requests for violent, sexual, or hateful content.

Cautious of potential misuse, OpenAI is also planning to introduce plans to adapt fake-image detection technology with Sora and will include industry-standard metadata tags to indicate how the content was generated.

Sam Gregory, executive director at Witness highlighed that the expressive capabilities of Sora offer new opportunities for storytelling but also raise concerns about misinformation and manipulation.

As generative video technology continues to advance, it will be crucial to strike a balance between innovation and responsibility, ensuring that AI-powered tools like Sora are used ethically and transparently in a rapidly evolving media landscape.