All I could think about as I used Sora, OpenAI's new AI social media app, was how much I missed the days of Vine. The good ole days, as I've come to think about them, when dumb internet trends were funny and, at a baseline, created by real humans for other real humans to enjoy. As I scroll past Sora AI videos, I can't help but lament that those days are long gone and may never return.
Named after its AI video generator, OpenAI's new app is custom-built to create and share AI-generated videos using its upgraded Sora 2 model. It's a big departure for the AI company, and it presents an interesting question to AI users: Would you use a TikTok-like app that's only populated with AI-generated videos?
I got my hands on a Sora invite code (the only way to access the app), and I am surprised to report it does feels like a social media app. Your main For You feed is an algorithmically curated, endless scroll of videos. You can like, comment on and share other people's videos, along with creating your own. You can filter between your friends and following feeds and, in a unique twist, filter videos according to your mood. But behind the familiar user interface lurks much deeper AI strangeness.
Don't miss any of our unbiased tech content and lab-based reviews. Add CNET as a preferred Google source.
Cameos, Sora's most unique and popular feature, allow you to generate videos using someone's likeness and place them into any kind of scene. Every Sora user can upload their cameo when they make an account and choose whether to let other people use their likeness in AI videos.
OpenAI CEO Sam Altman is undeniably one of the most cameoed creators on Sora in the short time the app has been available. I can't decide if it's a smart marketing ploy or the stupidest management decision ever made to let the general public make AI deepfake videos of Altman. Within 5 minutes of scrolling through my For You feed, I saw videos that cameoed Altman into the following situations: Getting arrested for stealing GPUs to run Sora 2, confessing that he cries himself to sleep because it feels good and stuffing his nostrils full of tissues while begging for likes. I made a video of Altman claiming Gemini was better than ChatGPT.
Watch this: Sam Altman's Deepfake Claims Gemini Is Better Than ChatGPT 00:12
Altman, presumably, has never done any of these things in real life. But the quality of the Sora 2 AI videos is so realistic that you wouldn't know otherwise.
Sora 2 excels at dialogue. That's a new addition to the model, and something we've seen AI enthusiasts appreciate, making it more useful for creators. Unlike with Google's Veo 3, you can get decent results without pasting an entire script in your prompt. You can say something like, "Altman sings a ballad about inference costs," and the AI will write the lyrics for you.
The upgraded model is also better at handling complex prompts, thanks to improved reasoning abilities. The videos take longer to generate, 2 to 5 minutes, likely because the higher-quality outputs take more time to process. While my attempts to get Sora to create videos featuring well-known celebrities like Taylor Swift were blocked by the moderation filter, others were able to get around them, like this video that features Altman in a field with Pikachu saying, "I hope Nintendo doesn't sue us." (Disclosure: Ziff Davis, CNET's parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
... continue reading