Hacker News Reader

Comment by pugworthy | Hacker News Reader

Why not feed all the descriptions from say, original Colossal Cave Adventure game into DALL-E 2?

"YOU ARE STANDING AT THE END OF A ROAD BEFORE A SMALL BRICK BUILDING. AROUND YOU IS A FOREST. A SMALL STREAM FLOWS OUT OF THE BUILDING AND DOWN A GULLY."

View on HN

Decided to try it out: https://labs.openai.com/s/nHmjLYmVPdDQzUU5DvGT2JgK

Wow, existing text-adventure could be very interesting with this kind of semi real-time illustrations. Maybe we'll finally know what is Grue like. Also, Table RPGs with Dall-E illustrations generated semi-realtime by the game master.

That was surprisingly good.

It's interesting that in both of the examples people have shared here, the image produced was quite good but entirely missed the "flowing out of the building" part of the prompt (which I think is a fairly significant aspect of the scene).

I picked the one that most included that; a few other suggestions didn’t have water at all.

https://labs.openai.com/s/DvQ49etAKKCU6Zpy32GFJlq0

The first thing I thought when reading this was "wow, someone must have fed Infocom games' location descriptions into an AI". Would still like to see that.

So frustrating that there's no way to access this tech without putting your name on a list and waiting an unspecified amount of time (months?). It's the sort of thing many people would pay well for, even at its current stage of development.

My wait was only 2-3 weeks FWIW.

I am still waiting since they opened the registration

I think it depends on what you tell them when you fill out the application. I was honest and didn't tell them I was a YouTube influencer, venture capitalist, underground artist, and/or credentialed researcher. I just wanted to play with it and see how it worked.

Most of the interesting things I've done in my life and career began with just wanting to play with something and see how it worked, but it was probably unrealistic to expect OpenAI to buy into that, given the large number of more worthy-sounding candidates in line ahead of me. Maybe I'll reapply with a different email address and make up a more mediagenic motivation.

I applied with the same criteria, and it took about a month to be accepted.

My invite actually came through just now, interestingly enough, after a little over two weeks. Either they're opening up for a wider audience, or whining about it on HN is irrationally effective.

Gotta give credit where it's due, in any case. I hereby withdraw my complaint!

Midjourney is open to everyone now, no waiting, and is easily the 2nd best tool for this stuff

Better than Stable Diffusion? I don't think so, especially in coherence.

Two replies are from Dall-E, copied here for convenience:

- https://labs.openai.com/s/nHmjLYmVPdDQzUU5DvGT2JgK

- https://labs.openai.com/s/DvQ49etAKKCU6Zpy32GFJlq0

For comparison, first two sets of 4 variations from MidJourney:

- https://i.imgur.com/iEvlYXE.jpg

- https://i.imgur.com/aR0R10p.jpg

The game with both is promptcraft. Here is Midjourney after changing “brick building” to “brick house” and changing “road” to “dirt road”. It did brick roads but at least a road showed up:

- https://i.imgur.com/bfia9zT.jpg

Then you can upsize to hallucinate additional detail:

- https://i.imgur.com/y1uPc6b.jpg

- https://i.imgur.com/LId4wy4.jpg

That actually sounds like a pretty interesting interactive art exhibition piece!

cool idea, but those images are so... un-text-adventure? I guess they feel like they should be more low res? I guess too "real" removes imagination from the text adventure equation

Reply on news.ycombinator.com