知Yizhi Daily
← Back to this edition

Conversations

Developer Uses Open-Source Models to Generate Isometric Pixel Map of the San Francisco Bay Area

Edition 2026-11-14 · 08:00 Beijing time

A developer showcased an isometric pixel-style map covering 1,300 square miles of the San Francisco Bay Area, using data from Google Photorealistic 3D Tiles. Users can locate landmarks such as Apple Park, the Palace of Fine Arts, and the University of California, Berkeley. The project reduced generation costs by training small open-source models.

The developer presented the project "Isometric SF" on Hacker News: it transforms about 1,300 square miles of Google Photorealistic 3D Tiles data for the San Francisco Bay Area into a uniformly styled isometric pixel map. Well-known landmarks such as Apple Park, the Palace of Fine Arts, and the University of California, Berkeley can be explored on the map.

According to the developer, the workflow first used Nano Banana 2 to train a Qwen-Image-Edit 2511 LoRA for prototyping the pixel style and generation method; the final map then used a Klein 4B LoRA trained with Qwen to reproduce the same style at roughly one-quarter of the cost. The developer said using Nano Banana 2 for the entire process would have cost approximately $7,000.

The developer said the experience was enjoyable for fine-tuning smaller open-source models and noted that the project was inspired by "Isometric NYC."

Original sources

Hacker News ↗
Back to this editionOpen interactive reader