OpenAI has unveiled “Sora,” its first experiment in creating custom models for video production from texts.
Through a series of updates and announcements on social media platforms, Sam Altman, one of the co-founders of OpenAI, revealed that the company has launched the new model and will provide access to it for a limited group of creative individuals starting today.
According to information posted on the OpenAI website, Sora is capable of producing video clips lasting up to a full minute, encompassing a wide range of artistic styles, including high-resolution and ultra-realistic facial shots.
The model can generate “rich scenes featuring a variety of characters, specific movements, as well as distinctive details related to the subject and background.” Moreover, Sora can take a single image and turn it into a video, or work on an existing video to extend its duration or complete missing frames.
Some flaws in the model, as mentioned by OpenAI, include difficulties in simulating complex environmental physics, lack of complete awareness of causes and consequences, as well as confusion in spatial details when requested, such as mixing up left and right.
Therefore, if your video includes a person running backwards, for example, it might be advisable to try reproducing it again. Additionally, this model is incapable of creating accompanying sound for the video.
Regarding how the company plans to prevent misuse of the technology, OpenAI announced its intention to include C2PA data, which would help track the “source” of the video and ensure it was generated by artificial intelligence when integrating the Sora application into any of OpenAI’s products.
Upon integrating Sora into one of OpenAI’s services, a text classifier will examine the provided data before execution to screen out any requests containing “gruesome violence, pornographic material, hate images, celebrity simulations, or intellectual property infringement.” After creating the video, another classifier will review it to ensure it is free of any prohibited content.
Among the “limited group of creatives” who will have priority to experience Sora are “visual artists, designers, and filmmakers” who will provide feedback to transform the model into a more useful tool for creative professionals.
Furthermore, researchers in the field of artificial intelligence will have access to this model to conduct penetration tests that involve uncovering errors and flaws, with the aim of developing and improving it.
Altman did not specify a specific date for making “Sora” available to the public. It can clearly be applied in business areas such as ad design, content generation for presentations, as well as content editing for social media platforms.
Altman asked social media followers to provide ideas for comments that could accompany the videos he creates, and announced his intention to design and publish these videos to give people a glimpse of Sora’s capabilities.
His first creation was a magical character wearing a pointed hat and a blue cloak patterned with white stars, casting a spell that releases sparks from his hand, while holding an ancient book in his other hand.
