Show HN: EarthToPixels – Turning real cities into explorable pixel art

EarthToPixels, a new project from a team including Tiago Trindade, turns real-world cities into explorable pixel-art maps by processing aerial imagery. The team fine-tuned an open-source Qwen image model and built a pipeline that generates pixel-art tiles from satellite photos, ensuring neighboring tiles remain visually continuous so they can be stitched into large, seamless maps, according to a post on Hacker News.

The pipeline was designed to work with arbitrary coordinates rather than a fixed location. For the first complete render, the team chose their hometown: João Pessoa, Brazil. The result is now explorable at earthtopixels.com, where users can browse the city as pixel art.

The project emerged from an effort to combine generative AI with nostalgic visual aesthetics. By leveraging a fine-tuned model and maintaining visual continuity across tile boundaries, the team solved a core technical challenge: preventing seams or jarring transitions when stitching satellite data into a cohesive pixelated map. The approach demonstrates how machine learning can transform geospatial data into new interactive formats.

This represents a niche but growing area where generative tools are being applied to maps and geographic visualization. The pixel-art treatment gives familiar urban layouts a retro, game-like quality while preserving the recognizable features and street layouts of real cities.