Browser Lab
Save a face as a local baseline, apply a modular AR intervention called a Ghostyle, and watch the face-matching pipeline react.
Ghostmaxxing in details.
Informative modules: grouped by the task you might want to complete.
Use the tools
Each guide follows the controls you can see, the result you can observe, the data that stays on the device, and the limits of the test.
Video Loader
Load an MP4 from your device, choose exact timestamps, record a baseline, and compare another frame without relying on a live webcam. Made to test make-up videos from popular sources.
Also inside the Lab: Face Brush lets you sketch a face-anchored digital intervention before attempting a physical makeup version. It is an optional authoring workflow, not a required first step.
Understand the results
Detection, landmarks, distance, and threshold answer different questions. The guides below connect those terms to the Lab without turning one local output into a general claim.
Read the Lab
Learn what the top readout means, why a missing face is not the same as a failed match, and what changes when a Ghostyle is active.
How face recognition works
Follow the pipeline from pixels to a face box, landmarks, an embedding, and a distance-based decision.
Glossary
Short definitions linked to the exact UI elements and implementation boundaries that matter in Ghostmaxxing.
Maintenance instruction and internals
If you want to understand the technology powering this lab, this is the right section.
Build and Share
Author a Ghostyle, inspect runtime modules and events, open the code map, or browse the generated JSDoc reference.
Maintain the project
Find every script in scripts-dev/, its command, inputs, outputs, overwrite behaviour, and known limits.
Generated reference
Module and symbol documentation generated from comments in the source code (via JSDoc)
Untile otherwise proven, any success is local/individual. A result depends on the model, camera, light, pose, distance, image quality, thresholds, and sample. A Ghostyle that changes this browser pipeline may not transfer to another model or a real deployment. Perhaps, once shared, repeated, tested under different condition, we'll have a reliable technique!