Openai decided to press on the table Shame 2. The proposal not only strives to dazzle with its technological progress, but also introduces a social format that makes anyone the main character of hyper -realistic clips ready for viralization.
He Jump relative to the first version Shara, launched in 2024, is obvious. According to the Compaa, the improvement is comparable to the fact that at the time was the passage of the GPT-3.5 to the GPT-4 in the text Mbit. Now objects and bodies support much more cunning physical consistency: the ball bounces on the board, not disappearing or teleporting, the characters retain the same condition from the same plane to the other, and the most difficult instructions are made with greater loyalty. The great novelty is that Sora 2 not only generates VDEO, but also integrates voices, effects and sound landscapes, opening the door to complete the works without the need for an external edition.
Social app for creating and sharing
A App that accompanies The model seeks to enhance this social code. After a brief face checking and voice, the user can be inserted as a coma into any scene, remixed by VDEOS and study the regulated natural language channel. According to Openai, the idea is not to contribute to an endless scroll, but to encourage the creation of a joint. The ambitious promise is that at the moment it opens in beta -phase in iOS with inviting access only in the US and Canada.
Examples Video SORO 2, run 10 minutes ago (via /@Openai) #Sora2
1/ man rides a horse on another horse pic.twitter.com/dp8rqa23xy
Fotachu – Ar Guy (@fotachuarguy) September 30, 2025
As expected, launch is accompanied by a package Security measures and digital well -being. Among them are automatic restrictions for minors, parental control elements, integrated in chat, options to withdraw the use of the image and supervision of human moderators to stop abuse. This is an obvious message to parents, educators and regulators: Openai wants to be different from other social platforms, where the uncontrolled exhibition and forage toxicity are periodically criticized.
Tcnic and visible liter of progress
The model, however, Not released from failures. In common demo, scenes, anime control, or researchers see in full snow, the results that amaze their likelihood, although the contradictions are maintained. Openai acknowledges that perfect is far away and there are many mistakes, but it insists that the failures now seem to be attached to the character of the stage, not with the algorithm. The nuance that can change the situation in the public.
Gymnasts Sora 2 for me very special pic.twitter.com/u1wrimfhip
Justin Moore (@venturetwins) October 1, 2025
More than technical, movement reveals Compaa Ambicin Under the guidance of Sam Altman: to turn the shame into a new cultural epicenter, as it happened with Chatgpt in 2022 or with Studio Ghibli images in 2023. When Bet works, VDEOS will stop being only a consumer product to convert to a common game space where anyone can become an actor, director and screenwriter in a few seconds.
(Tagstotranslate) false tags
