Welcome to a special Wednesday edition of the AI For Humans newsletter!

Before we get into Sora 2… our early beta for AndThen is now open! More information below but if you just want to try what we’re working on:

Go here to try the (early) beta of AndThen.

Play the conversation. It’s weird and fun.

Sora 2 is VERY Fun & Surprising

Yesterday was a big day for OpenAI. Sora 2 is here. 

Their brand new state-of-the-art video model appeared semi-suddenly with their classic livestream tweet & it’s pretty good! And, surprise, it’s also an entirely new social network. Kind of.

It’s mostly a great AI video model that generates with realistic sound & physics plus a few special features. For a quick example of what it can do, check out this video I made which has gone semi-viral within the Sora app.

It’s honestly hard to believe how long ago the Sora 1 announcement was (February 2024 for those counting). It was a ground-shifting moment not only in the AI space but in Hollywood as well.

It was the first time that AI video was taken as a serious competitor to what Hollywood might be able to make and, while those early videos didn’t fully come to fruition in the release of Sora 1, Sora 2 seems like it’s opening the door to a different sort of idea, one of social-first AI video people might actually make.

But first…

What Sora 2 Does Well & How It’s Different

There are few important things to know about Sora 2 that set it apart:

Sound: Sora 2 videos have sound much like VEO 3 & it’s the first time that an OpenAI video model has been released with this function. In our tests, it performs about as well (if not slightly worse) than Google’s VEO 3 but overall, it’s good enough.

Listen to the below for a quick sense of how it is with squeaking sneakers…

Physics: Sora 2 is actually MUCH better at physics than most other video models, at least at first glance. I need to do a deep dive comparing it to the latest Kling model but you can already tell that it’s a large step up over what VEO 3 has been capable of.

Cameo: The underlying secret that powers Sora 2 is that it’s not just an AI video model but also it allows you to upload your face & voice to their servers (if you’re so inclined) and allow you to use your persona to power AI video outputs.

This sort of tool is in use in MANY other AI video platforms but the implementation here is pretty slick and, in my limited testing, very powerful.

The capture process is simple and done within the app and will 100% make personalized AI video MUCH more popular. Once you upload your Cameo, other users can @ your name to include you in AI gens (if you’ve given permission). Again, it’s very fun and I can see a ton of friend groups getting into this.

We’ve yet to see if Cameo will cause a full-blown Ghibli-style moment for OpenAI but it does make the other aspect of what they launched today much more compelling than I thought it would be at first…

Sora 2 Uses AI Video To Power A “Social” Network

Do we really need another AI social network? After all, the Meta announcement from last week went over like a sack of bricks.

But what separates OpenAI’s new Sora app is the social factor.

Because friends can make videos with other friends’ Cameos, that opens up an entirely new world of people that might use this. It also give users a reason to go to this site vs just a place to create and consume the dreaded AI slop.

I’m actually feeling much more positive the Cameo aspect alone might make a difference for the success of this as social network. If nothing else, it means that the content will feel a lot more personalized than AI video has before.

Of course, there is the thorny issue of IP rights and I’ve already seen a TON of content that feels like it’s unlikely to last on the app for long. The WSJ broke a story a few days ago that OpenAI gave companies the abilty to opt-out of Sora but this example feels like we’re likely to see a few people still upset.

How can you get access to Sora?

Well, it’s at the invite-only stage for now. If you know someone who’s already in, they’ve got five invites to send out. It’s a smart way of staggering who gets on when and will prob help them manage server load.

The good news is that, for now, it’s entirely free to use for those that are in the app. This won’t last forever (neither will the IP-laden outputs) so if you’ve got someone you can beg for an invite, let them hear it.

I’d give you one of ours but I’m already out!

But, more importantly, before we go…

Our New Thing: The Early Beta of AndThen

Kevin & I have been working on something for a bit and we’re excited to share with you… the early beta of AndThen! 

AndThen is a new platform for real-time voice interactions with AI characters. Humans create the scenarios, backstories & rules of each experience and our AI model brings them to life. 

Each time you play an AndThen experience, our system responds directly to you based on what what you say to it. You steer the conversation with your voice and it responds. And this isn’t idle chat, each of these experiences has something for you, the human, to do, a goal to accomplish.

The big key for us here is that, while these are driven by AI models, humans create the scenarios behind them. Human creativity will continue to drive compelling narratives forward and we believe that some in the AI space are undervaluing that part of the creation loop.

It’s something we’re really proud of and think there’s clear path to finding ways to make something new with AI + Humans at the same time.

We’re starting off with a handful of experiences you can try. Some of them will update daily (our game show Say Less!) and some are more one-off experiences.

We’re excited to hear what you think. Pop into our Discord and we’ll try to respond as quickly as we can.

Also, if you want to make one of these, we’re working on a full-blown platform and we’d love to hear from you. Scroll to the bottom of the AndThen website for how to get in touch with us.

More very soon (like literally tomorrow)!

-Gavin (and Kevin)