> The source is Google Photorealistic 3D Tiles. Isometric.nyc explored and rejected the use of 3d building data. It is pretty insane that US gov has free LIDAR data for every city in the US available to the public. I spent <30mins exploring this and stuck to google 3d images. Claude Code whipped up a scraper to stream the 3D Tiles and render with three.js. This gives the best "real" texture base for the model to learn from. Anything else would involve a LOT of manual work to get the inputs right.
> Now that we have the 3D tiles, we need to generate the ground truth pairs for training! Similar to isometric.nyc, I generated a few ghibli-style pixel-art images using Google's Nano Banana. SF terrain is very interesting. There are quite a few distinct features like skyscrapers in FiDi, the hills in the southeast, 2 iconic bridges, lots of coastline, piers, parks, suburban grids and lots of water. I generated a ton of images and curated from them. Getting consistent style was a challenge. There was a LOT of manual trial and error. But as usual, Claude Code added this feature to the dev app that allowed me to select the best images and approve them.
RugnirViking 2 hours ago [-]
interesting with the alcatraz island - I notice its invented some houses, and turned a cruise ship into a rock. I appreciate the difficulty and getting a lora to understand it - despite this the finished project is beautiful
I also feel that some of the technical explanations are a little dense here in the dev blog. I forget who it was who I saw mentioning it here on HN, but I've been having a lot of success with getting models to (re)write technical explanations in https://en.wikipedia.org/wiki/Simplified_Technical_English
simonw 11 hours ago [-]
> I'm stoked for the golden era of side-projects!
Love that attitude. Me too!
nuwandavek 9 hours ago [-]
very inspired by you!
nuwandavek 9 hours ago [-]
glad you liked it!
I had to leave out a ton of details in the training process. But at every step, codex/cc built me a dev tool to navigate the complexity. That was such a unique side project experience.
11 hours ago [-]
rob74 6 hours ago [-]
Nice! But for a city renowned for its steep streets, this looks suspiciously flat :) Ok, you could say it's only a 3D rendering of a 2D map, and it probably has to be this way for practical reasons, to avoid hills obscuring objects behind them, because you can't rotate the map as far as I can see, but not seeing actual height represented in a 3D view still takes some getting used to.
Also, there are neighbourhoods and landmarks, but not actual street names as far as I can see? Not to mention transit stops/lines and other fancy stuff one could add...
RugnirViking 2 hours ago [-]
from his dev blog: "All map math happens in a single fixed ... plane... The intended effect was a classic Sim City view of the city. "
sgt 54 minutes ago [-]
Sim City 2000 for example had elevation [1]. The issue is that LLM's simply don't "get this". It's vibecoded.
It's vibe coded. LLM's have a hard time understanding elevation and isometric display.
notpushkin 4 hours ago [-]
It’s vibe coded but it’s not slop. There’s a lot of human input here. So either the author overlooked the elevation issue or it was a deliberate decision, but in any case you can’t just pin this on “LLMs don’t understand isometric 3D”.
sgt 55 minutes ago [-]
I've been busy with a rendered isometric map recently and I figured this is something Opus can help with, but it was near useless at understanding isometric designs along with elevation changes. It kept me busy for days.
donatj 43 minutes ago [-]
I've spent about a month all told in SF over a couple visits.
As a devout midwesterner, I didn't think I was going to enjoy the place. Having spent a bit of time there however, and having been shown around by natives, I absolutely get it.
There's so much to see and do packed in a small area, and the air just feels electric.
"Satellite images from highly oblique angles are pretty mindblowing"
If u like this u will like that
nuwandavek 10 hours ago [-]
woahhh, these are so beautiful!
murphyslab 12 hours ago [-]
Incredible. It's easy to just go on browsing and exploring. The massive, scrollable pixel art aspect reminds me a little of Floor796: https://floor796.com/
tamimio 12 hours ago [-]
This is amazing, it got everything I could think of but not all, where’s little nightmares, silent hill, naruto, men in black, for a starter. But it’s neat regardless!
The author continues adding and refining the project, so if it isn't in there yet, perhaps in the future.
tamimio 11 hours ago [-]
Nice, love it!
smusamashah 11 hours ago [-]
i remember seein men in black here
codetiger 7 hours ago [-]
This is incredibly beautiful.
A few months back I tried to build a my own version of Google Earth rendered in voxels. Basically building from the terrain heightmaps scans available in open public data. Making it smooth and tessellation to work was incredible amount of work even for a simple vexed renderer
https://codetiger.in/earth/
tda 6 hours ago [-]
Nice, but not sure I agree with your use of voxels. It appears more like pixelated height map, not the use of 3d pixels. From what I see every position on the globe can only have a single elevation
codetiger 5 hours ago [-]
Each pixel is a cube drawn at a certain height based on the terrain height map data. It’s not a volumetric render, I agree.
jtfrench 12 hours ago [-]
Those who know game dev know making good isometric maps can be deceptively difficult. You grabbed that bull by the horns and did so beautifully. Well done!
nuwandavek 9 hours ago [-]
ty!
cyanregiment 10 hours ago [-]
A few AI anomalies, like several roads turning into rivers and lakes.
There are also a few massive square ponds in the Tenderloin that I am pretty sure do not exist lol
I suppose that explains the yellow filter and the mixels
accrual 12 hours ago [-]
This is really cool, it gives me SimCity 2000 vibes. I like that I can click on neighborhoods and regions and get more information about them!
yreg 7 hours ago [-]
I'm currently playing with something a bit similar and I was wondering about the fair use aspect.
What's the consensus on sharing the stylized assets created by diffusion models based on reference photos from e.g. Google Earth? Can the final output be shared as public domain or do Google, Airbus, et al. still hold some rights over them?
jcarrano 6 hours ago [-]
The same "fair use" as when Google trained its model on everyone else's photos.
qurren 10 hours ago [-]
Would be extra awesome if there was a navigation app that used this map
This could be a great anecdotal benchmark for agent/model progress. I wonder how much easier things have gotten with the latest Codex/Claude vs. when isometric.nyc was made.
flomo 10 hours ago [-]
Just one little thing: East of Buena Vista Park, you have a lake, it is actually the road which goes around the park.
Also, I would love to zoom in more. (pls give me the pixels)
nuwandavek 9 hours ago [-]
Spurious lakes are an interesting issue I looked into a bit. 2 types of cases ended up becoming lakes:
1. "shadows"
2. flat green areas
I think a better data distribution should eliminate this. Also, my "water detector" was not too good and could be improved.
Very cool! Yesterday my wife and I watched Inside Out, and seeing the San Francisco setting felt very familiar (last year, 2025). Last year, in 2025, my wife and I went to San Francisco for our honeymoon. This website also showed me many familiar places.
montag 12 hours ago [-]
Really cool. I can see some issues, like Starr King park turned into a lake for some reason, but it’s so much fun to look at. Do you have any way to patch errors and discontinuities at the tile boundaries?
I tried out a few ideas, but i think better training data around water will help a LOT.
toplinesoftsys 10 hours ago [-]
Fantastic project! Sim-city over San Francisco.
harrisonpage 13 hours ago [-]
Amazing! I was sad I couldn't zoom in more tho
nuwandavek 10 hours ago [-]
oooof, I can actually allow more zoom (was just embarrassed at bad pixels)
nuwandavek 9 hours ago [-]
more zoom enabled!
greatony 13 hours ago [-]
THIS IS SICK!
The BGM, the SFX, the visual are all perfect.
My favorite is the banner plane, this is really really SF
nuwandavek 10 hours ago [-]
haha ty!
burnoutdv 5 hours ago [-]
I am not a fan of this.
Like the new york one, its just...unpolished. I dare to say slop. I am sure the author put some effort into it, BUT its just not there for me. The wiggly lines, the blurry designs and just everything. With something is pixelart or appears to be, I expected it to be pixel perfect, not just some random picture that looks like pixelart on the first glance.
The idea feels neat, but the execution needs so much more touch and polish. Maybe even handcrafting to be impressive.
hasyimibhar 7 hours ago [-]
Sorry but that is not pixel art.
jullls 5 hours ago [-]
Reminds me "transport tycoon", very cool!
quemtrouxe 3 hours ago [-]
it's pretty interesting
danenania 12 hours ago [-]
Super cool! Can you talk a bit about how you built it?
Thank you for putting this together. Brought some fun memories back from my time living in SF
https://sf.isopolis.city/dev.html
> The source is Google Photorealistic 3D Tiles. Isometric.nyc explored and rejected the use of 3d building data. It is pretty insane that US gov has free LIDAR data for every city in the US available to the public. I spent <30mins exploring this and stuck to google 3d images. Claude Code whipped up a scraper to stream the 3D Tiles and render with three.js. This gives the best "real" texture base for the model to learn from. Anything else would involve a LOT of manual work to get the inputs right.
> Now that we have the 3D tiles, we need to generate the ground truth pairs for training! Similar to isometric.nyc, I generated a few ghibli-style pixel-art images using Google's Nano Banana. SF terrain is very interesting. There are quite a few distinct features like skyscrapers in FiDi, the hills in the southeast, 2 iconic bridges, lots of coastline, piers, parks, suburban grids and lots of water. I generated a ton of images and curated from them. Getting consistent style was a challenge. There was a LOT of manual trial and error. But as usual, Claude Code added this feature to the dev app that allowed me to select the best images and approve them.
I also feel that some of the technical explanations are a little dense here in the dev blog. I forget who it was who I saw mentioning it here on HN, but I've been having a lot of success with getting models to (re)write technical explanations in https://en.wikipedia.org/wiki/Simplified_Technical_English
Love that attitude. Me too!
I had to leave out a ton of details in the training process. But at every step, codex/cc built me a dev tool to navigate the complexity. That was such a unique side project experience.
Also, there are neighbourhoods and landmarks, but not actual street names as far as I can see? Not to mention transit stops/lines and other fancy stuff one could add...
[1] https://patcoston.com/sc2k/city/images/Calebopolis_Big.png
As a devout midwesterner, I didn't think I was going to enjoy the place. Having spent a bit of time there however, and having been shown around by natives, I absolutely get it.
There's so much to see and do packed in a small area, and the air just feels electric.
It's genuinely a different world.
"Satellite images from highly oblique angles are pretty mindblowing"
If u like this u will like that
Naruto: https://floor796.com/#t2l4,684,675
The author continues adding and refining the project, so if it isn't in there yet, perhaps in the future.
There are also a few massive square ponds in the Tenderloin that I am pretty sure do not exist lol
But I like the idea.
Planning to fix in the next version!
I suppose that explains the yellow filter and the mixels
What's the consensus on sharing the stylized assets created by diffusion models based on reference photos from e.g. Google Earth? Can the final output be shared as public domain or do Google, Airbus, et al. still hold some rights over them?
This could be a great anecdotal benchmark for agent/model progress. I wonder how much easier things have gotten with the latest Codex/Claude vs. when isometric.nyc was made.
Also, I would love to zoom in more. (pls give me the pixels)
I think a better data distribution should eliminate this. Also, my "water detector" was not too good and could be improved.
I wrote about a few of these things here: https://sf.isopolis.city/dev.html
I tried out a few ideas, but i think better training data around water will help a LOT.
Like the new york one, its just...unpolished. I dare to say slop. I am sure the author put some effort into it, BUT its just not there for me. The wiggly lines, the blurry designs and just everything. With something is pixelart or appears to be, I expected it to be pixel perfect, not just some random picture that looks like pixelart on the first glance.
The idea feels neat, but the execution needs so much more touch and polish. Maybe even handcrafting to be impressive.