ChatGPT 4o Adds Image Generation and It’s Fun as Hell

MG Siegler:

I’m not even talking about Apple and AI here. We’ve done that, a lot. Probably enough — for now (famous last words). I’m talking about Apple in general. Watching this OpenAI video — again, not an event, just a product walk-through with various team members (though this one happened to be “MC’d” by Sam Altman) — I had this old, familiar feeling as they walked through the new features: joy.

As ridiculous as it may sound, I was almost giddy around what I was seeing. It’s a feeling that I recall well from many an Apple event back in the day.

“Yes, this is exactly what I wanted! They did it!” That kind of thing.

This will sound unfair or harsh to Apple, but I really don’t think that it is. I can really only speak for myself here, and perhaps I’m alone — but I suspect that I’m not — it has been a while since I’ve gotten that feeling from an Apple announcement. That loving feeling.

I gave the updated ChatGPT the instruction “Create an image of the main characters from ‘Severance’ as Lego figures” and this is the first response it gave me:

Four Lego figures who look like Mark, Helly, Dylan (maybe?) and Irving (even less maybe).

I gave Apple’s Image Playground, running on MacOS 15.3.2 Sequoia, the equivalent prompt — “The main characters from ‘Severance’ as Lego figures” — and it gave me this as its first response:

Three Lego figures, two yellow, one red, which bear zero resemblance to anyone or anything from “Severance” whatsoever.

One of the above images qualifies as “Hey, that could be better but it’s pretty good for the first crack from a simple prompt”, and the other qualifies as “This bears zero resemblance to anyone or anything from Severance”.

Guess which one of the two Apple is actively promoting to users as something they should try?

Friday, 28 March 2025