Welcome back to the AI For Humans newsletter!

I’m writing this from San Francisco as we prep to present our new startup AndThen to a bunch of investors as part of a16z Speedrun’s Demo Day.

Go here to try AndThen and let us know what you think in our Discord.

But now, on to the Sora 2 discourse to end all discourses…

As Expected, the World Freaked Out About Sora 2

We gave you all the details last Wednesday about OpenAI’s Sora 2 release and, wow, we might’ve actually undersold how big a deal it would be.

From all accounts, this launch has been a smashing success for OpenAI.

It’s on par, if not bigger, than the launch of their 4o Image Gen model (remember Ghibli-gate?) and, it seems to me, MUCH larger than GPT-5 for the mainstream public. And, unlike that one guy’s product, it’s now the number one app in Apple’s App Store.

The sheer amount of both a) headlines and b) social conversation it has generated feels unprecedented.

Of course, a huge part of that conversation is around the controversial decision to release the tool with a LOT of IP baked into the model.

According to reports, OpenAI did give rights holders a week or so to opt out ahead of time but it’s clear that either some of those holders did not get the memo (Nintendo is not the most communicative company) or maybe some of them thought it wasn’t going to be that big of a deal.

One particular thing I don’t think many people thought was how good this model would be at writing and talking like some of these characters. And, while it’s one thing to see South Park characters generated in silent videos or photos, it’s an entirely different thing to see what very nearly could pass as a full blown South Park episode.

But After Three Days, a Different Sora 2 Emerges…

That pressure has been felt.

There was a huge course correction on Friday as OpenAI’s Sam Altman very quickly shifted tacts and pulled a lot of the access to third-party IP.

Some rightsholders might have been “excited” about this but it’s more likely a lot of them were wondering just how they would keep control over what people would do with their characters. And, starting Friday, it became a LOT more difficult to make nearly anything.

Honestly, I found myself feeling strange about a few of the things that I saw as well.

AI People Are Big Mad About Sora 2 Too…

Something I didn’t really expect is that there are quite a few people from inside the AI industry who are not happy with Sora 2 either.

There was Anthropic, which has begun what looks like an ‘anti-slop’ campaign and then Hard Fork’s Kevin Roose had a lot to say about how this might not exactly align with standards OpenAI hopes to uphold.

I think it’s worth reposting something here we discussed on last week’s show where Sam responds to some of these critics…

I tend to agree here that this is fun overall and that there’s something remarkable about what they’ve unlocked with Cameo. Never before have people been able to open multiversal portals with their friends like this. And, yes, as an avid scroller (and enjoyer!) of TikTok, I’ve found the Sora feed to be quite joy inducing at times.

But I also think this is a big deal and it’s hard to overstate what this means for both our personal identities and Hollywood going forward.

The Dawn Of A New Social Age… and More?

Can you tell I’ve been semi-obsessed with this whole thing?

Like what it means to… us?

And when I say ‘us’ here, I’m not talking about Kevin & me or even you, dear reader.

I am literally talking about what it means to us as a species.

It sounds hyperbolic but I think this moment is one of the first where we kind of start to feel the impact of AI across what it means to be human. We’ve been toying with these weird chatbots and, yes, they keep getting smarter but something about this changes a bit of the way we see reality going forward.

I will say it’s worth watching Casey Neistat’s video about it. I found it an interesting argument even if I don’t totally agree.

That’s it for today. See you on Friday for the podcast!

- Gavin (and Kevin)

In this week’s AI For Humans: Our OpenAI Sora 2 First Takes & Best Usecases👇

We 💛 This: Sora 2 Prompting Best Practices

Of course, we couldn’t let this week go by without sharing some of the best Sora 2 prompting tips we’ve seen over the last week.

Something that’s already very clear is that you can create a list of shots and Sora 2 is pretty good at following them. A few days ago, on my way to LAX, I created the following ‘day-in-the-life’ sort of post & it did ok. This is a just first output and I’m sure if I threw this into the machine a few times it could potentially improve it a bunch.

A few more great tips we’ve seen so far below:

• Using the Remix feature to allow you to make longer videos (that you can then assemble yourself for posting)

• Pixel Art workflow step-by-step: Nanobanana → Sora 2

• Another Nanobanana → Sora 2 step-by-step to make meme images (and a bit of an end-around on Sora 2’s no real people uploads)

• Don’t forget it’s VERY good at archival footage (prob a lot of training data) and this is a fun B&W prompt you can use.