AI created this

May 13, 2026 at 7:03pm

This is getting kind of scary.

I made this myself in like 30 seconds using the new GPT image generator. Prompt was "Maplewood New Jersey 1908 real estate promotion pamphlet." I used Nightcafe, which is an AI image generator service that includes a variety of different image generator models. This was the fastest lowest quality GPT model.

Then I made this map. Don't use it to navigate but ... still pretty amazing.


The images are higher res before uploading, my version doesn't show them so fuzzy. They look pretty real.


It does seem to have some navigational failings.  grin

That seems odd to me. Can't AI access Google Maps?


That isn't how AI image generation works. It creates configurations of pixels that are similar to other maps it has in its database. What seems new is that it is integrating its text-based knowledge base into the image. It is also incorporating more specific visual imagery than I have seen before.  E.g. Maplewood village looks similar to the real thing (except that light pole in the middle of the road might be a tad bit inconvenient).

Given how hyper specific it is, though, I think eventually it will even get the map right. Just five years ago a similar request for a map resulted in the below.

And I also find it really scary that it created a fake yet plausible (on first glance) historical document. The world is going to be swimming in this sh-- before you know it.


HatsOff said:

Then I made this map. Don't use it to navigate but ... still pretty amazing.

what was the prompt you used for the map?

I use some of the other bots but I haven't really used chatgpt much. I wonder if gemini, copilot or claude have similar functionality.


I'm sure I could ask one of them.  cheese


HatsOff said:

That isn't how AI image generation works. It creates configurations of pixels that are similar to other maps it has in its database. What seems new is that it is integrating its text-based knowledge base into the image. It is also incorporating more specific visual imagery than I have seen before.  E.g. Maplewood village looks similar to the real thing (except that light pole in the middle of the road might be a tad bit inconvenient).

Given how hyper specific it is, though, I think eventually it will even get the map right. Just five years ago a similar request for a map resulted in the below.

And I also find it really scary that it created a fake yet plausible (on first glance) historical document. The world is going to be swimming in this sh-- before you know it.

I think the race is for the A.I. to be better trained before "A.I. slop" so infects what's available online, that it starts to infect the results of A.I. prompts.

For example, if too many "don't-use-it-to-navigate" maps are online, they can affect how the A.I. constructs a map in response to a request for one. I don't know how they're going to put more "discernment" into the large language models, although I know they're trying.


drummerboy said:

what was the prompt you used for the map?

I use some of the other bots but I haven't really used chatgpt much. I wonder if gemini, copilot or claude have similar functionality.

"Maplewood NJ 2026 illustrated map." The older one was "city road map detailed." It never would have occurred to me back then to give a more specific prompt.

To answer your other question, Claude doesn't do graphics other than very simple schematics. Not sure about the others - I fooled around with copilot a while ago and wasn't impressed, but much has changed since then. 

Probably not super relevant but I didn't use ChatGPT. it's the same engine ChatGPT uses though.

I don't do a lot of this but I try to test stuff when it comes out so that I have a passing familiarity with what is going on in AI. 


nohero said:

I think the race is for the A.I. to be better trained before "A.I. slop" so infects what's available online, that it starts to infect the results of A.I. prompts.

For example, if too many "don't-use-it-to-navigate" maps are online, they can affect how the A.I. constructs a map in response to a request for one. I don't know how they're going to put more "discernment" into the large language models, although I know they're trying.

absolutely. Already when I look at Pinterest half the things I see are obviously AI generated. If you do image searches at places like stock.adobe.com you can filter AI images out and when you do you notice just how many things there are AI. 

The thing is ... it's getting harder to tell the difference.


Try asking ChatGPT to run it through Google Gemini to validate streets. Yeah, they can bounce prompts back and forth. 



In order to add a comment – you must Join this community – Click here to do so.

Sponsored Business

Find Business

Rentals