Hey, I'm the solo developer behind this project. Here's the story: I got frustrated with the traditional remote sensing workflow. If you want to detect "cars," you train a CNN on a cars dataset. Need to find "blue shipping containers" next week? Start over with new data and a new model. The fixed-class bottleneck is brutal. So I built a pipeline that uses Vision-Language Models (VLMs) to skip that entirely. You just type what you want to find, and the AI searches the satellite map tile-by-tile. The pipeline is simple: Pick a zoom level + type a prompt → each map tile gets fed to the VLM → bounding boxes are created → projected to real geographic coordinates → drawn on the map. What works well: - Distinct structures (bridges, stadiums, swimming pools, cars) are detected reliably zero-shot - Relational prompts like "cars parked on driveway", "boats on land" The free demo requires zero login, just draw a polygon and type a prompt: https://www.useful-ai-tools.com/... I'd love to hear what objects you try to find! Happy to answer any questions about the tech stack or approach!

Detect Anything: Satellite Analysis

Find any object in satellite imagery with AI

Find any object in satellite imagery with AI