LaughingMan

Overlays the Laughing Man logo from Ghost in the Shell: Stand Alone Complex onto faces in a webcam or video stream — a native Rust app that composites a multi-channel-SDF logo on the GPU.

Overview

A capture → detect → track → composite pipeline: a frame source (webcam, video file, or a synthetic test pattern) is scanned for faces, and the logo is drawn over each face by a Vulkan compositor. The logo is rendered from baked MTSDF atlases so it stays crisp at any size and its text ring can spin independently. The overlay stays level with the camera (it does not tilt with the head).

Build

Requires a recent stable Rust toolchain.

cargo build --release

Optional feature flags (off by default to keep the runtime build light):

Feature	Enables
detect	The SCRFD ONNX face detector (pulls ONNX Runtime). Without it, a synthetic demo face drives the overlay.
bake	The laughing-bake tool + the mask module (pulls fdsm/usvg/resvg). Needed only to bake the logo atlases.

Quick start (no logo, no camera)

Runs a synthetic scrolling test pattern with a procedural placeholder logo — nothing to install:

cargo run --release -- --test-pattern

Preparing the logo

The real Laughing Man art is copyrighted and not included — supply your own laugh.svg. The logo is composited as three layers, each a separate SVG baked into its own atlas (all in one shared frame so they stay aligned):

Layer	Contents	Purpose
--static	Rings, face features, cap, text band — everything except the rotating text	The still logo + the white face backing (derived at load)
--text	The circular quote only	Spins around the logo's centre
--front	Features + cap	Its silhouette occludes the ring so the cap reads as in front of the spinning text

Split your laugh.svg into these three layer SVGs — each keeping only its own part (a vector editor works, or delete the paths that don't belong to a layer). Name them laugh-static.svg, laugh-text.svg, and laugh-front.svg, then bake them (the text layer is the slow one — thousands of glyph segments):

cargo run --release --features bake --bin laughing-bake -- \
    --static laugh-static.svg --text laugh-text.svg --front laugh-front.svg

This writes assets/static.mtsdf.png, assets/text.mtsdf.png, and assets/front.mtsdf.png (--out-dir defaults to assets, --size to 1024, --range to 8). A single combined SVG can also be baked with --svg <file>.

Running with the logo

cargo run --release -- \
    --logo-static assets/static.mtsdf.png \
    --logo-text   assets/text.mtsdf.png \
    --logo-front  assets/front.mtsdf.png

• --logo-front is optional; omit it and the spinning text draws over the cap instead of behind it.
• With only --logo-static (no --logo-text), a single combined atlas is composited without a spinning ring.
• With no logo flags at all, a procedural placeholder is used.

Controls

Key	Action
S	Save a full-resolution screenshot (video + overlay) to screenshots/laugh_<timestamp>.png
Esc / Space	Quit

Frame sources & options

Flag	Effect
(default)	Capture from webcam device --source <index> (default 0)
--video <file>	Decode an mp4/mkv/… as the source (ffmpeg is auto-downloaded on first use)
--test-pattern	Synthetic scrolling pattern, no camera
--size WxH	Requested capture resolution (e.g. 1280x720); defaults to the camera's highest
--list	List capture devices and exit
--probe	Print capture dimensions + measured rate to stdout, no window

Face detection

Detection uses an SCRFD *_kps ONNX model and the detect feature. Without a model, a synthetic demo face positions the overlay:

cargo run --release --features detect -- --model scrfd_500m_kps.onnx --video clip.mp4 \
    --logo-static assets/static.mtsdf.png --logo-text assets/text.mtsdf.png --logo-front assets/front.mtsdf.png

Report Issues

Submit an issue

Contribute

Submit a Pull Request!

License

Licensed under the MIT License.