From Photo to Graphite: The Geometry Behind Realistic Drawing Animation

Why Traditional Filters Fail at Drawing Animation

Most video editing software attempts to simulate drawing using linear wipe masks, radial reveals, or simple opacity dissolves. While fast to render, these methods do not replicate how human hands draw:

Humans do not draw from left to right: An artist completes discrete semantic objects (outlines, eye contours, hair segments, hatching) before moving across the canvas.

Humans draw along vector tangents: The angle of the pencil or stylus aligns dynamically with the direction of the stroke.

Strokes have varying speed: Artists accelerate on long smooth curves and decelerate at sharp vertices.

---

The 5-Stage Speedpaint Trajectory Architecture

Stage 1: Frequency Decomposition

The input photograph is processed through edge-preserving bilateral filters and neural line decomposition to separate structural contours from soft tonal gradients.

Stage 2: 1-Pixel Skeletonization

Using topological thinning (Zhang-Suen algorithm), thick contours are converted into exact 1-pixel wide mathematical centerlines without breaking junction topology.

Stage 3: Semantic Object Clustering

Rather than randomly visiting line fragments, strokes are clustered by spatial proximity and semantic connectivity: 1. Primary Structural Outlines: The foundational silhouette. 2. Focal Anchor Details: Critical facial features, mechanical edges, or architectural frames. 3. Interior Contours: Secondary line work that defines form. 4. Hatching & Tonal Cross-strokes: Density-based reveal passes.

Stage 4: Minimum-Jump Path Optimization

An internal travelling-salesperson path graph minimizes the "air jump" distance between pencil stroke endpoints, creating smooth and natural transitions.

Stage 5: Reveal Radius & Tangent Slant

During video compositing, each stroke point dynamically dictates the pencil tip's angle, pressure modulation, and graphite reveal radius on the canvas surface.