After the presentation in February of this year, OpenAI released the final version of the AI model for text-to-video generation — Sora. It is reported that the service is available to those who have a paid ChatGPT Plus or Pro subscription. However, it is noted that even after the release, users still have to wait in a long queue.
According to the company's press release, users will be able to create videos with a resolution of up to 1080p and a duration of no more than 20 seconds. However, the quantity and quality of these videos will depend on the type of subscription chosen by the user.
The ChatGPT Plus subscription allows users to create up to 50 videos in 480p and 720p resolution, but with a smaller quantity. Pro account holders will be able to create up to 500 videos in 1080p. OpenAI reports that starting in 2025, they plan to introduce "individual pricing" for different user categories. Those using the free version of ChatGPT will always be able to view videos, but will not have the ability to create them.
The music for the video was created using Suno, and the video was generated in Sora. Both the music and video are inspired by a cyberpunk theme, and creating all of the content from scratch took 6 hours.
OpenAI also noted that the model has its limitations. For example, some generated videos feature unrealistic physical properties of objects. The goal of the public release is to gather feedback and collaborate on developing standards and norms for this technology. Each video created with Sora contains a C2PA "watermark." Currently, the company is restricting certain types of content, including materials with explicit sexual content. However, in the future, some of these may become available if effective measures against deepfakes are implemented.
It is also worth noting some of the main user complaints from those who have already formed their opinions after using Sora.
Users have pointed out that Sora often fails to fulfill text requests accurately. For example, instead of generating three pyramids, it created only one. When asked to depict five people, the model showed only two, and it failed to generate a sphinx altogether. In one instance, where a cat was expected to appear on a chair, Sora could not place the animal on the furniture despite four attempts.
When creating 20-second videos, Sora shows unpredictability: it either creates abrupt cuts with frequent angle changes or slow-motion scenes that do not meet user expectations.
Working with context also presents difficulties. Even to create realistic videos, such as one about Ancient Rome, users had to make multiple attempts to achieve the desired result.
The release of Sora also lists some other limitations. For example, Sora is not intended for individuals under the age of 18, is not included in the Team, Enterprise, or Edu subscription plans, and is unavailable in the UK, Switzerland, and EEA countries.