Camera Lucida

A Humble AR Assistant

It started with a simple problem: expensive optical camera lucida tools and clunky tripods were getting in the way of learning to draw. We wondered if the incredible spatial awareness of modern devices like iOS and Apple Vision Pro could pin a reference image natively into our physical environment. The first commit was just a basic image tracker—a proof of concept to see if AR could replace physical tracing paper.

Beer Glasses and Desk Rigs

Software is only half the battle. We needed a way to hold the phone perfectly still over a sketchbook without spending money. We discovered the ultimate indie-dev solution: balancing the iPhone horizontally across a tall pint glass. The glass provides a perfectly stable, overhead stand that overlooks the sketchpad without getting in the way of your arms while drawing!

Value and Chiaroscuro

As we started using the app, we realized that drawing isn't just about tracing outlines—it's about understanding light. We dove into classical art theory and implemented the Global Value Snapshot and Monochrome filters. By mathematically breaking down the reference into its core values, we built tools to help us see the world the way the Old Masters did.

Color Harmony

Once we understood value, we tackled color. We built a mathematical Color Harmony engine that analyzes the RGB pixels of the reference image and extracts the dominant color palettes. This allows artists to visually isolate hue relationships and pre-mix their physical paints on their palette before ever touching the canvas.

Combating VIO Drift

We hit a major technical roadblock. During a 20-minute tracing session with the phone stationary on a glass, the image would slowly drift inches across the paper. ARKit's Visual Inertial Odometry (VIO) accumulates noise when staring at blank paper. To fix this, we invented the Hybrid Anchor Transplant.

Instead of relying on a static world anchor, when the user locks the image, we reverse-raycast to find the physical table (ARPlaneAnchor) and transplant the image tracking root. To prevent the image from snapping violently during the transfer, we explicitly solve the offset matrix:

let requiredLocalTransform = planeAnchor.transform.inverse * entityWorldTransform
entity.transform = Transform(matrix: requiredLocalTransform)

This completely eliminated the drift, providing absolute millimeter precision for tracing.

Contours, Mirrors & Foreshortening

With tracking locked down, we added powerful overlays to simulate historical optical phenomena. We integrated Canny edge detection to automatically generate contour lines for tracing. We also implemented Convex and Concave Mirror Distortion filters to simulate the exact optical distortions studied in the Hockney-Falco thesis, alongside 1, 2, and 3-point Radial Perspective Matrices.

visionOS and Beyond

We brought the app to Apple Vision Pro for true spatial immersion. Most recently, we spent time tearing out clunky UI paradigms (like the UIKit ARCoachingOverlayView and inline SwiftUI Menus) and replacing them with lightning-fast native Action Sheets to ensure the creative process is never interrupted by software lag.