I want to talk today about Sora 2 and the new Sora AI app.
First of all, Sora is not new. OpenAI released Sora about a year ago initially and, at the time, it was a sort of hidden part of ChatGPT that let you make bad videos. And trust me, they were bad. I tried, hard, to create something usable that I could maybe use as like a cutscene or some interesting visual teaser for the Heckna D&D mini-campaign I was running at the time. While it was interesting, there was nothing even remotely usable.
Even things that started off from real images got weird in unrealistic ways. It was just an experiment in how badly AI does video.
Based on the AI-generated videos that were generated from that time, my experience was not unique. One of my favorite viral videos was a social media video of two Chinese kids performing a live action reproduction of AI-generated slop videos and it was (and still is, although the context has now shifted) brilliant.
Sora 2 is not that.
Sora 2 is able to produce 9-second videos that look and feel real with a minimum of weird AI hallucinations. There are far fewer random limbs sprouting from weird body parts (although, not zero (content warning on linked video: disturbing body stuff)), the AI “actors” look like real captured video, and you can even upload a video of yourself to add yourself (or your friends, or anyone who allows their likeness to be shared) to your own AI generated videos.
Why would you use AI to put you into a video starring yourself when you could just literally take a video? Lots of reasons. Maybe you want to show yourself skydiving but you’re terrified of heights, or maybe you want to show yourself flying away into a tornado, or turned into a K-pop star, or giving a news report from inside the literal internet. Or maybe you just want a Ring cam video of Sam Altman being attacked by squirrels. (Each one of those examples is an actual thing I’ve either seen or done myself, with the exception of the Sam Altman video, but I’m sure now having typed it out, that someone will create a video with exactly that prompt.)
Never-ending supply of yummy slop
A lot of people are going to talk about the quality of AI video and its potential for deep fakes. And yes, that’s definitely there. But I expected that and this post isn’t about that. Instead, I want to talk about something I did not expect: how completely, mind-numbingly addictive this stuff is.
I will admit, I have not partaken of the addictiveness of TikTok. And, maybe that’s not relevant. I certainly won’t claim that the “algorithm” is somehow intuiting what Sora videos I’m more interested in vs. not. I don’t think it’s that sophisticated. But I don’t think it needs to be. Because, before I knew it, an hour had passed of me just scrolling through (and making my own) weird AI-generated videos and I didn’t realize it.
Not all of it is good. In fact, a lot of it is bad. Not bad in an offensive way, just bad in an uninteresting way. People making some of the popular creators and YouTubers scream or say stupid things is not particularly interesting to me. Endless remixes of MLK’s “I have a dream” speech where he’s calling for everything from a lowering of OpenAI’s content restriction policy to telling Andrew to use Sora is occasionally amusing.
And that’s sort of the thing. Once a video is out there, others can “remix” it. A “remix” takes the original video and adds a new prompt to it, often simply “this but with x” — x in this case could be replacing the person in the video, adding an element or changing the dialogue. Remixes appear in line with the original source, so you can sort of swipe through endless remixes of a particular video and see how it evolves over time. (This experience only exists in the mobile app. The desktop experience does not (yet) have a way to scroll through remixes.)
A great example is a video someone made of a guy scraping a frozen windshield with a credit card, making a 15% off joke. Then the implement he uses to scrape the windshield changes to various different things. Then it becomes a blowtorch. Then he’s grilling burgers on his frozen car. Then he’s grilling burgers on an actual grill and falling backwards onto the fence behind him. Then he’s grilling burgers but he has no utensils and is using his hands to flip the burgers. It’s sort of like a game of exquisite corpse, each person adding their own little touch until the end result becomes something totally different.
Honestly, watching the evolution of these 9 second videos is probably the biggest draw for me, and the thing that makes it most addictive. When anyone can make a high quality clip that basically rivals Hollywood, the creativity involved in one-upping the last one, or putting your own unique stamp on it makes the experience interesting. Plus, there’s that little shot of dopamine every time you hit something new. And it’s almost impossible not to run into something new. There’s a enough of a steady flow of of this slop to make me concerned how long before we as a global population start boiling the oceans with the amount of computing power we need to run the data centers for these things.
It also can’t be understated how easy it is to make a stupid 9 second video that’s at least somewhat entertaining (and if it’s not entertaining enough, well, you can tweak the prompt to make it better). It didn’t take very long before I created a whole wall of unused draft videos with something resembling my face in almost all of them. AI has gotten my face wrong before, in pretty upsetting ways. There’s a special kind of uncanny valley reserved for seeing yourself the way an AI thinks you look. But Sora manages to do a pretty decent job of it, which is both upsetting and impressive.
I’m sure you’ll be seeing a lot more Sora content — both content about Sora and content generated by Sora. I don’t exactly know how to feel about it all yet. But I do think that part of what makes the app experience and scrolling through these stupid AI brain rot videos worth it is that there is an underlying human creativity behind all of it, even in its most vulgar or idiotic. There are people making the AI generate these stupid videos, and some of them are actually pretty clever.
[this post was created with a lot of AI-generated content, but none of the words were written by AI.]
Leave a Reply