OpenAI has finally launched the next generation of its text-to-video generation model, Sora 2, along with a new social media app that is powered by the new tool. Sora will compete directly with Google’s Veo-3, which became an instant hit among the masses after being launched to the public earlier in the year.
“Sora 2 can do things that are exceptionally difficult—and in some instances outright impossible—for prior video generation models,” OpenAI said in a blog post.
OpenAI says that Sora 2 is better at obeying the laws of physics, which is an important capability for any video generation model. For instance, the company says that with the previous generation of video generation, if a basketball player missed a shot, the ball may spontaneously teleport to the hoop, but with Sora 2, the ball will rebound off the backboard.
The new model is also said to be better at controllability, and OpenAI says it can follow intricate instructions spanning multiple shots while accurately persisting the world state. The model is touted to excel at realistic, cinematic, and anime styles.
Just like Veo-3, Sora 2 can also create background soundscapes, speeches, and sound effects which go with the flow of the video, bringing a degree of realism to the scenario.
One interesting fact about Sora 2 is that it is capable of inserting humans or any other real object directly into the created video.
“The model is far from perfect and makes plenty of mistakes, but it is validation that further scaling up neural networks on video data will bring us closer to simulating reality,” OpenAI says in its blog post.
OpenAI’s New Sora App:
The company also seems to be looking to chase another viral moment with Sora 2’s “upload yourself” feature and is launching a dedicated social media app called Sora on iOS.
OpenAI says users will be able to “create, remix each other’s generations, discover new videos in a customizable Sora feed, and bring yourself or your friends in via cameos.”
The cameos feature would allow users to enter any scene on the app after a one-time video and audio recording in the app.
OpenAI says in order to minimize doomscrolling and addiction with its new app, it will give users the ability to be in control of their feed. The company has developed a new recommendation algorithm that can be instructed through natural language.
The ChatGPT maker also says that by default it shows content “heavily biased towards people you follow or interact with, and prioritize videos that the model thinks you’re most likely to use as inspiration for your own creations.”
