His first presentation surprised everyone. Sora, OpenAI’s artificial intelligence capable of creating videos from simple text, is now available to the general public. The first lucky people to have tested the tool testified to the power of AI. And to worry about the future of video creation.
The first feedback, notably that of the famous YouTuber MKBHD (who was able to use the final version a week before everyone else) reveals impressive capabilities. The tool particularly excels at creating abstract animations, simulating fluid effects like water or fire, and generating cartoon-style sequences.
Sora, how does it work?
The interface is intended to be minimalist and accessible: a simple text box allows the user to describe the desired video. Customization options include choice of resolution (from 360p to 1080p) and duration (up to 10 seconds). Generation time varies depending on the complexity of the project, ranging from a few seconds to several minutes.
One of Sora’s major innovations lies in its integrated storyboard system. This feature allows you to chain several prompts together to create more elaborate sequences. We can thus circumvent the limitation of the tool to manage complex scenarios in a single request.
A particularly interesting feature allows you to remix existing videos. Users can modify them by adding or changing specific elements while preserving the overall feel of the original sequence.
Some improvements to be expected
Despite his technical prowess, Sora is not without his faults. Movements, particularly those of human characters, can sometimes appear unnatural. Object permanence also needs improvement: certain elements tend to disappear or transform inconsistently throughout sequences.
Ethically, OpenAI has built in many protections. The tool categorically refuses to generate content featuring minors, public figures or registered brands. Each video is also marked with a watermark, but this protection can be easily circumvented by a simple crop.
There also remains the painful question of the data used to “train” Sora. OpenAI discusses the use of “publicly accessible media”but no precise details on the nature of this data and the withdrawal possibilities for the original creators were given.
Available to everyone… or almost
Like many artificial intelligences recently unveiled, Sora is available today in the United States. European users will still have to wait. OpenAI has not given an official explanation for this delayed launch.
To use Sora, it is possible to try using a VPN to locate yourself in the United States. But nothing says that OpenAI has not developed protection against this practice. We will know more in the coming hours.
- OpenAI officially launches Sora, its video generation AI capable of creating realistic sequences from textual descriptions
- The tool offers varied options (resolution up to 1080p, durations up to 10 seconds) and excels in abstract animations and special effects
- Launch delayed in Europe
