Alright, Codex with GPT 5.5 is completely cracked. This is nuts 🤯 Basically one-shotted my request to create an app that takes a prompt, creates an equirectangular panorama with GPT image 2 – and then use Apple's ML Sharp to stitch a gaussian splat world together.

First turn produced a working app that does what it says on the tin. A few more turns created what you see here. ML Sharp didn't like the equirectangular image, so GPT 5.5 split it up, and then spliced multiple PLYs together. Clearly a prototype, but still – bonkers.

Don't forget to follow for more vibe-coding experiments.

Alright, well a lot of you are here so I guess I'll plug my new app! It's a fun satisfying sculpting app now on iPad and iOS! Go check it out.

If you're interested in AI, you might enjoy this video I made on The Connection Machine. The AI Supercomputer of the 1980s that was 30 years ahead of its time. If you like it, subscribe – I have more on the way!

That’s awesome! I saw someone else play with equirectangular images with ChatGPT images 2 and my mind went straight to what you just built!

And this was literally like, 5 mins 🤯 Obviously it's a crude prototype — but, being able to just ask for it to wire together all these separate pieces is so valuable for ideation.

I love that we can always tell it's GPT by the gradiated HTML dropdowns. Lol

I know, it's like hold up Claude can you take a pass at this UI codex made real quick 😂

The one-shot capability is wild. I've been building automation tools and the jump from "describe what you want" to "here's working code that actually integrates multiple APIs" is massive. Apple's ML Sharp integration especially - that would've taken weeks to prototype before.

Exactly — I wanted to see how far it could go in terms of fitting many pieces together that would normally be a pain. From setting up the image API, to grabbing the various libraries and making them talk to each other, etc.

GPT 5.5 aside, that’s actually a really slick app idea ! well done !

Ooh, amazing. Is this app itself on GitHub or somewhere. I have a use for something like this.

Thanks! No, this was a five minute little experiment so I have no plans to post it to github right now — but the prompt was more or less what I said in the original post! I bet you could give this to 5.5 and ask it to create the same starting point.

In the past, we had to construct a world; now, we need only describe a dream. Codex seamlessly fuses the "visual power" of image generation with the "spatial power" of Gaussian Splatting. By 2026, code will become the sole medium connecting imagination to physical space. This is the Singularity.

@neurofoo GPT-5.5 Codex one-shotting Worldforge is straight-up sorcery 🤯 We’re so back in the prompt-to-3D era!

This is super cool use case

Imagine when 5.5 Pro is available in Codex. It will be fun

the gaussian splat part is sending me wtf

The future is moving way too fast and I'm absolutely here for it 🔥

That is actually impressive! Definitely stepped up

@OpenAIDevs Outstanding work right there..

I can has app?

Holy moly fried ravioli

Mamma Mia!

Can this be an Apple Vision Pro app?

Oh that's sick!

Now everyone is Fei Fei from World Labs and Google Deep Mind with Genie 👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀👀

Holy moly

crazy times.

I'm not sure "one-shotted" is doing the work here. Did it actually ship something usable or did it get you 80% there and leave you debugging the last 20% for two hours?

Impressive

@ChatGPTapp 5.5 is beyond next level!!!

@gil_n_ Love it

Integrating GPT 5.5 and image 2 directly into Codex has boosted productivity up to a new level. Feels like heavy Game Scene Modeling could be easily done now.

I didn’t understand that prompt but was impressed nonetheless.

That is the interesting benchmark now. One shot demos are nice, but the real question is whether GPT 5.5 keeps that level once the task gets messy, needs tools, and has to recover from its own mistakes mid run.

That is the kind of demo that makes the model jump feel real. One shot app generation is impressive, but the bigger unlock is when people can trust the second and third iteration too without the whole thing turning brittle.

@OpenAIDevs WHAT

one-shotting splat worlds is the new hello world

Yeah, this is really really good actually useful total step change. I knew right away. Based on the response. People are going to take a couple days to see how good it really is once the Smart people get involved.

The part that stands out is the end-to-end flow. Once the model can handle setup, tool glue, and retries in one loop, “prototype” starts turning into real product speed.

Exactly; quality can be iterated on. But connecting the dots saves so much time to get from idea to prototype.