Currently OpenAI is hosting a 12-day livestream eventEarlier, there were reports that Sora, which many have been waiting for, would be officially released around this time, though the exact timing was unknown. Just a moment ago, on Reddit and X, people began leaking clips of videos generated by Sora v2 shown in an internal meeting. Not only was the video quality astonishing, comparable to Hollywood, but they also claimed it would be coming next week.
Suspected Sora v2 leaked video surfaces, made up of multiple short clips spliced together
OpenAI first unveiled its text-to-video AI model Sora on February 15 this year, allowing users to generate AI videos featuring multiple characters, different camera movements, and background details from brief or detailed text descriptions, or even static images. At the time, OpenAI promised that it would definitely launch this year.
However, there was no news of a release afterward, and I originally thought it would be delayed. But with the leaks and OpenAI’s 12-day livestream event, hope gradually returned—and it even directly leaked the Sora v2 version. Since v1 hasn’t even been released, it’s a direct jump to v2.
earlier Reddit ForumOn the OpenAI discussion board, a user shared a leaked 1-minute Sora v2 video, in which it’s clearly evident that there’s a major improvement over previously released videos. Many clips could be called Hollywood-level. The characters’ movements, eye contact, and expressions are all very natural, like the war scene at the beginning; Sora can now also generate multi-person battles. The author also wrote that Sora v2 will arrive at Christmas.
Sora v2 Leak – 1-Min Video Output, Image-to-Video, Video-to-Video, and more. Coming Christmas.
byu/Designer-Pair5773 inOpenAI
Meanwhile, on the X.com platform, a user named Siqi Chen also shared the same clip, and the timing he leaked is even more specific: it will debut next week, that is, before December 20:
leaked footage from openai's sora 2.
i thought this kind of quality was maybe a year away but here we are.
coming next week.pic.twitter.com/Os3Y4JBqkx
— Siqi Chen (@blader) December 8, 2024
OpenAI’s 12-day livestream event has completed 2 days so far, which were last Thursday and Friday. That means there are still 10 days left. Excluding the upcoming weekend, the livestream event will continue until next Friday. If Sora is truly launched during this event, there’s a high chance it will appear as the grand finale, so we can look forward to December 20.
According to current leaks, Sora v2 is expected to generate 1-minute videos, offering three modes: text-to-video, text + reference image to video, and text + video to video.
Although the leaked footage is quite impressive, each clip is too short, which has led many netizens to doubt and criticize it. For example, one commenter said: “These clips are all 3 to 4 seconds long. Like other existing models, there is currently no way to extend or control the duration. This clearly requires a lot of post-production effort.”
Also, the funny thing is that these models can generate fairly realistic-looking people, but they absolutely cannot edit these two-second clips into a logical story. The Vikings start off in fierce combat, then stare at each other, put on a depressed expression, and only then board the ship???
To impress me, show a continuous 60-second scene with no camera cuts and consistent content. These results so far don’t look any better than what hailouai and kling already have.
Source: KOCPC Chinese