Openai has presented Sora 2, the improved version of its video generation model with artificial intelligence. The company has been working on new capacities for the tool, which now includes audio and more realistic and precise videos, as well as in a new experience that materializes in a social media application, opening the era of synthetic social networks.
Sora’s original model was presented in February 2024 and was officially launched last December, with a first version that, despite the limitations, impressed to the market. Now, the model has trained to offer creations that resemble real world more. The company considers that Sora 2 could be to the generation of video what GPT-3.5 supposed for the chatbot.
This is described from OpenAi: “Sora 2 can do exceptionally difficult things, and in some cases completely impossible, for previous video generation models“They comment in a statement.”The previous video models are too optimistic: they transform objects and deform reality to successfully execute a text instruction. Although it is still imperfect, it better obeys the laws of physics compared to the previous systems”
Illustrates it with an example. With Sora if a video was created in which a basketball player fails a shot, the ball can teleport spontaneously to the hoop. Instead, with Sora 2, if a basketball player fails a shot, bounces on the board. In this sense, the company indicates that they have perfected a useful capacity for every simulator, which is to be able to model failure, not just success.
The new version of Sora not only offers video, but is now capable of creating sophisticated sound landscapes, dialogues and sound effects with a high degree of realism. In addition, it allows to integrate elements of the real world, such as animals, objects or even people, to achieve simulations with a precise representation of appearance and voice.
A social network of Deepfakes
The company accompanies the presentation of Sora 2 with a new social application, called “Sora” and available for iOS systems under invitation in the United States and Canada. With it, users can create, discover new videos in a customizable feed or include themselves or their friends in videos. Openai explains that users can appear directly on any Sora scene, which calls cameos, “After a brief video and audio recording in the app to verify your identity and capture your image”
The application resembles Tiktok, as can be seen in the Apple Store. It has a “for you” page and an interface with vertical displacement. Includes the function called “Cameos”, which allows users to give permission to the app to generate videos with their image. You can also remember the image (even when interacting with that of other people) describing the desired video and audio in a text instruction.
All this raises a new scenario for the generation of simulations and deepfakes. Openai, however, presents it from the perspective of entertainment and creativity. “We believe that people can have a lot of fun with the models we are building”, They say in the statement.”We started experiencing with this “Over” function several months ago in Sora’s team, and we all have a lot of fun. It seemed a natural evolution of communication: from text messages to emojis, voice notes and this“
Developed to avoid the doomscrolling
Openai ensures that he has developed the application according to the risk of doomscrollingaddiction, or isolation. In this regard, he points out that he offers users tools and options to control what they see in their feed, and that will periodically probe users about their well -being.
It also explains that by default, it shows content with a strong bias towards the people who are followed or with which it interacts, and the videos that the model considers that it is more likely that the user will use as inspiration for their own creations are prioritized. “We do not optimize the time you spend in the feed, and we explicitly design the application to maximize creation, not consumption”, They say from OpenAi.
On the other hand, he says he is establishing predetermined limits for the number of generations that adolescents can see a day in the feed, and implementing stricter permissions for cameos. It also claims to be expanding human moderator teams to quickly review cases of harassment if they arise and establishing parental controls.
With regard to the cameos, OpenAi points out that the user has total control of his image. In this sense, only the user decides who can use their cameo and you can revoke access or eliminate any video that includes it at any time. You can also see the videos that contain their cameos, including drafts created by other people, at any time.
In addition, according to it The Vergeeach video created with Sora will include signals to indicate that it is generated by AI, such as metadata or a water mark. The company also says it has internal detection tools to help evaluate whether a certain video or audio was created with Sora. The application also does not allow screen recording.
With regard to the Deepfakes, Openai has explained, as reported by the aforementioned medium, that the image of public figures in Sora cannot be generated unless they have uploaded a cameo and given their consent for its use. Nor will it be possible to generate content for adults or “extreme” through the platform.
However, the precision and sophistication achieved by Sora 2 increases the difficulty in distinguishing what is real from what not, and Openai’s incursion into the field of social media passionate the path to a synthetic reality. Despite this, the company considers that its proposal will improve the existing. “We see this as the beginning of a completely new era for cooperative experiences. We are optimistic and we believe that this will be a healthier platform for entertainment and creativity compared to what is currently available”They comment.
They also predict a new era of creativity and cultural development. “The general world purpose simulators and robotic agents will radically transform society and accelerate human progress. Sora 2 represents a significant advance towards that goal”, They point out from the company.“We believe that Sora will bring a lot of joy, creativity and connection to the world”
