gnipbao/whiteboard-video-engine

Whiteboard Video Engine is a Python command-line tool that takes a photo, illustration, SVG, or line drawing and converts it into a video where the image is drawn stroke by stroke, like someone drawing on a whiteboard. The output is an MP4 file. The tool is designed for local use and focuses on the rendering mechanics rather than cloud processing. The core pipeline runs in several stages. A local neural network model first extracts line art from the input image, supporting realistic photos and anime-style illustrations via two different model backends. The extracted strokes are then sorted and smoothed, and a simulated hand gesture follows along each path as the lines are drawn. After the line work is complete, color can optionally be added using a contour-based fill effect. FFmpeg handles the final video encoding. The hand overlay comes in four built-in gesture styles (asian, black, children, white) or can be disabled entirely. Stroke detail level is configurable (balanced, rich, max) and the tool auto-detects appropriate line thickness from the input. A --draw-text flag converts a short title string into a handwritten path that gets drawn into the video. Installing the tool requires Python and FFmpeg. The line art models (Informative Drawings for photos, Anime2Sketch for illustrations) are optional but improve results with non-SVG inputs. Each model backend needs to be set up as a separate project directory with downloaded weights, then pointed to via environment variables. A "whiteboard doctor" command checks whether all dependencies are in place. The license is MIT. Model code and weights from the upstream projects keep their own separate licenses.