Skip to content
Shelved · 2025

Orbital Survey Visuals

Turn a photograph into a publication-quality scientific visualization paired with AI-generated abstract art - delivered as a single self-contained HTML file.

  • Python
  • NumPy
  • SciPy
  • scikit-learn
  • HTML
  • Claude API
What it was

Every output is built to look like it belongs in a Nature journal figure or a gallery print: precise computer-vision overlays rendered at 300+ DPI, paired with a unique abstract artwork derived from the image's own data. Python handles the vision analysis; Claude makes the per-image design decisions and generates the art.

I really liked the outputs. They felt very sci-fi and offered a very nice decomposition of the photographs - luminance contours, HSV palettes, etc. Plus, it was my first time getting to use Anthropic's API in an external app.

The sub-project Artemis 2 is a separate take on the organic-art idea: instead of Claude-generated SVG, it runs a real particle-flow simulation. It builds a flow field from image gradients, seeds hundreds of thousands of particles biased toward bright/edge regions, and simulates them. Then it HDR tone-maps and blooms into a luminous 4K PNG - Refik Anadol-style generative art, with Claude setting the art-direction parameters.

Scientific visualization of a drone landscape photo with luminance contour lines and a measurement grid.
Contour-mapped overview - luminance topography and a feature grid pulled from a single drone frame.
Drone photograph overlaid with a triangulated mesh network and an HSV colour wheel.
Delaunay mesh and HSV-palette decomposition of the source photograph.
Drone photograph with a point-network overlay and an extracted colour palette.
Point-network overlay with the image's dominant colour palette.
Why it didn't stick

It's not completely killed, just shelved and not actively used. Maybe when I take more photos and dissect them, I'll iterate on it.

Learned about extrapolating data from images and free-flow particle simulations in chaotic mode.