Highlights:

  • The newly launched version of Sora, Sora Turbo, features enhancements that allow it to generate videos much faster.
  • Users can access Sora via a newly launched dedicated website, which includes various tools designed to simplify the video generation process.

OpenAI has officially launched its highly anticipated Sora video generation model for ChatGPT Plus and Pro users.

This innovative algorithm first appeared in February as a preview release. The newly introduced Sora Turbo, version of Sora, includes enhancements that enable it to generate videos significantly faster.

ChatGPT Plus subscribers can generate up to 50 videos per month, with a maximum resolution of 720p and a length of up to five seconds. In contrast, users of the recently launched ChatGPT Pro plan, priced at ten times more, can generate up to 500 videos per month. Pro users can create clips up to 20 seconds in length, with a maximum resolution of 1080p.

Users can access Sora via a newly launched dedicated website, featuring an interface with various tools aimed at streamlining the video generation process.

A video project begins with a prompt where the user outlines what the clip should depict. Users can personalize the style of the generated frames, adjust the length of the clip, and tweak other settings. The model generates the video in one of three aspect ratios: widescreen, vertical, or square.

OpenAI has equipped Sora with the ability to switch between aspect ratios by training it on “spacetime patches.” These data units are akin to tokens, the information snippets that a large language model processes as text. Spacetime patches offer a standardized method for storing the multimodal data handled by a video generation AI.

Just as tokens can store various types of text, such as prose and code, spacetime patches can store videos in different aspect ratios. OpenAI developed these patches for training Sora through a two-step process. The system converted each video in the training dataset into a latent space—an abstract mathematical representation that takes up less storage than the original file. It then divided the latent space into smaller segments, with each segment representing an individual spacetime patch.

The technology offers additional advantages beyond enabling Sora to adjust video aspect ratios. According to OpenAI, using spacetime patches allowed it to train Sora on videos with varying durations, resolutions, and aspect ratios, simplifying the development process.

In addition to Sora’s aspect ratio settings, the company provides a range of advanced controls for further customizing videos.

Rather than using a single prompt to create a clip, advanced users can divide the video into segments and apply a unique set of instructions to each one. If a frame doesn’t meet their expectations, they can adjust it by submitting a follow-up prompt. Additionally, Sora provides the option to extract a frame and expand it to create an entirely new video.

A feature known as Blend allows users to merge two clips into a new video. In another section of the Sora interface, the Featured and Recent feeds display videos created by other users.

“OpenAI’s launch of Sora marks a transformative moment in AI-generated video technology. While it opens many doors for creative content, it resurfaces already pressing questions around generative AI about copyright, authenticity and the future of creative industries. As AI capabilities continue to evolve, it remains crucial to establish proper regulations, tools and overall best practices that safeguard authenticity and ensure ethical use in this rapidly changing landscape,” Alon Yamin, Co-founder and Chief Executive of AI-based text analysis platform Copyleaks said in an email.

The original version of Sora, previewed by OpenAI in February, could generate clips up to one minute in length. With the current limit set at 20 seconds, it’s likely that OpenAI will update ChatGPT in the future to allow longer videos.

Sora may soon be introduced in the business versions of ChatGPT, which currently do not include this model. If OpenAI decides to integrate Sora into these plans, they might also implement features tailored specifically for professional video teams. For instance, the company could introduce a shared content library that enables teams to store their Sora-generated assets in one central location.