Sora App: Forget about plot-based media!
The new Sora app is quite impressive and has large implications across AI, social media, & entertainment.
I recently heard an account of Gen Alpha scornfully refer to “plot-based media”, seemingly any media not found on scrollable phone feeds. The scorn implied plot-based media wastes your time with formulaic plots and takes forever to get to the point. And while plot-based media has dominated mainstream culture, a totally plotless media based on vibes, memes, and sound bites has thrived on the Internet where you can watch algorithmic feeds of crowdsourced strangers mimicking online trends.
This dichotomy, while incomplete, is useful for understanding the scattered media landscape.
The Legacy Establishment (movies, television, & books) pursues plot-based media.
The Internet pursues vibes, memes, and what we might call feed media.
In the last two years, there’s been much consternation over AI invading the first category. But increasingly, it feels like AI may pass over Hollywood and the traditional entertainment industrial complex. AI is useful around the edges of production, but video models are clumsy tools for long-form storytelling. AI is much better suited to the second category. With the release of the Sora app, it leaves no doubt in my mind. OpenAI has created a platform that leans into the natural strengths of AI models and it really works.
In this column, I’ll write about the new Sora app. I’ll cover:
What it is
Why it’s fun
and 16 reactions about it
The Details
Sora is OpenAI’s new TikTok-like video app for creating and watching AI-generated videos.
You can create videos by describing them in plain english. When you prompt a video, Sora generates dialogue, audio (voices, music, sound-effects) and video as a single (mostly) coherent output. Under the hood Sora likely significant meta-prompting to automatically expand users prompt like “clowns on bikes” into very descriptive prompts with enriched details. For example, if you prompt a ‘South Park episode about Sora’, it very impressively infers and executes the intended visual style, voice-acting, music, and even writing style.
In the feed, you scroll through an endless supply of AI-generated videos made by other uses on the app. Many videos contain people you know (your friends) and things you know (SpongeBob, Mario, Ronald McDonald).
You can also remix any video by simply describing the changes you want like “add Spongebob to the video” or “change it to be all in Spanish”. This creates a new version of the video with your desired changes. Again, the meta-prompting allows you to write commands as simple as “anime”, and Sora will remix the video into an anime with the appropriate visual style, voice-acting, music, and more. For any video on the platform, you can scroll right and see every remix of it. Some videos have seemingly infinite remixes.
Optionally, you can scan your face and voice to create a ‘cameo’. This allows you to put yourself into videos. You can also create videos of any friend who’s also scanned their likeness. The app offers controls around who can use your cameo and how you’d like to appear.
It’s very easy to create in Sora. You can be sub-literate and still create fun videos. If you see something you like, you can remix it with the command “funnier” or “make fat” or even just “again”. Any of those prompts will return a watchable video.
Sora has removed a lot of the ‘creative friction’. It’s really easy to make entertaining videos. I’ve found there is minimal payoff for trying hard. In fact, the less hard you try the more fun you have. It’s more about entertaining yourself and your friends than realizing a pre-existing vision.
The Fun
The Sora app is particularly addicting and fun.
Sora offers the familiar addiction of short-form video feeds along with a social network and notifications about friends making videos with your likeness. There are new unclear social norms to test as well. Is it uncouth to cameo with a married woman? Can you offend someone with their own AI avatar? How should you feel that someone made your avatar do that?
Almost immediately I made videos of acquaintances doing bizarre things. Then I made videos of me slapping a friend in increasingly surreal ways. In both cases they seemed to think it was funny. Obviously lines will get crossed here sooner rather than later.
Generating videos is fast, but not instantaneous. The ~2-minute video generation time lags behind your imagination and your excitement. I found myself queueing up multiple ideas and constantly checking my phone to see how they turned out.
And importantly, no video ever turns out as you planned. If you have a specific intention, you’ll almost always find it betrayed, butchered, and lost in the output. So you’ll try again. And again. And again. There is such sheer volume of videos (people post dozens of videos a day), even your favorite videos are quickly lost in the pile.
The point of Sora is not the content. It’s the act of making it. Media creation is fully recreational here.
My Thoughts
Sora is an incredible app. It leaves me with the following thoughts:
Unlike every previous AI video app, Sora is not a means to an end. It is an end in itself.
The goal of Sora is not to create entertainment for disinterested viewers. The purpose is to entertain yourself and your friends. It’s peer-to-peer media.
Peer-to-peer media is a huge shift away from media that’s made by specialists as a product to be consumed by a wide audience.
Up to this point AI video apps have chased the plot-based media segment, marketing themselves to “filmmakers” and “artists” with an emphasis on human-in-the-loop creation.
In Sora you do not need any special skill or understanding to make entertaining videos. The point of Sora is to offer the easiest possible video creation. There’s significant meta-prompting happening so that any user input is expanded into a high quality Sora prompt that will produce a good output. I expect this will only improve with time.
Meta-prompting is a technique of using prompts that write prompts during the generation process. It seems trivial for OpenAI to automate Sora’s entire feed. ChatGPT could write 1 million unique user prompts that are turned into 1 million AI-generated videos. It could then write 2 million more prompts to remix those videos. And etc…
The human-in-the-loop model may disappear sooner than I thought. I’m not the only one surprised by the speed of technical progress here. A very smart friend (he helped build Veo 3) told me, “Candidly I feel like this human in the loop of creation is going to die way sooner than I thought. Like I think we are only going to get closer to one click generation.” For feed media, I do not disagree with him. For plot-based media, I disagree.
I’ve never watched myself so much! I don’t get photographed often and I am rarely filmed. In 48 hours on Sora, I ‘watched myself’ more than I ever had previously. That was very odd. I think it improved my self-image actually.
The “cameo restrictions” make a point of saying “only you can see preferences set here”. People will try to shape perception of themselves. I already saw one account of a woman saying she had re-recorded her cameo with makeup on to appear more attractive.
Users can set preferences for how their likeness is used in Sora videos. There are also celebrities on the platform. You can create a video with Jake Paul or Sam Altman. Many celebrities will surely opt-out of Sora. Will there be a weird viral fame for those who opt in? There were probably more videos of Sam Altman created in the last 7 days than ever previously existed.
The speed of meme-ing is accelerating.
The platform is clearly unlike any social media or entertainment app that’s ever existed before. It will have large effects on our self-image and our minds.
The app is also attached to ChatGPT and uses your data. It’s unclear what this means. Will OpenAI launch AI avatars with your likeness trained on your ChatGPT data? Will it spin up a meta-verse based on what it knows about you?
The app is suspiciously hard to delete. You can’t delete it without also deleting your ChatGPT account too. Not only that, if you delete your Sora account you are forbidden from using the same email or phone number to create a new account.
Sam Altman really doesn’t want you to delete Sora... What type of user data is OpenAI gathering from Sora to train the next model?
Why did OpenAI make this? Why would OpenAI build a social app designed to entertain you? Only a few years ago Sam Altman was trying to scan everyone’s eyeball for World Coin. People called him weird. Now everyone is voluntarily scanning themselves into his social video app.
I have lots of questions about Sora and how people will use it. What’s unquestionable is Sora is a Totally New Thing. It’s blown the door off AI video. Overnight it’s given us a real, entertaining use case for AI video that normal people will enjoy without needing tutorials or caveats.
I’m learning to never underestimate OpenAI. You never know what they’re going to do next. But it might be wise to start thinking about it.
Super interesting about the difficulty deleting 🤔 also the preferences made me laugh 🤭
I too am growing tired of plot based media, as it turns out??