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/
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?
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!
> 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 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.
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
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.
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.
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/
59 comments
[ 0.24 ms ] story [ 18.9 ms ] threadI tried out a few ideas, but i think better training data around water will help a LOT.
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.
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.
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
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)
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.
"Satellite images from highly oblique angles are pretty mindblowing"
If u like this u will like that