Quickly integrate face, hand, and/or pose tracking to your frontend projects in a snap β¨π
Powered by:
Handsfree.js.Portfolio.mp4
Explore examples: https://labofoz.notion.site/901ff075d8df4bd7b9500d134c22038b?v=f33d13f1c67b458487250992a8501e2c
Hello! I was recently displaced from my home following a few traumatic events. I wasn't able to raise enough support and so I deleted my socials and am just going to work through this on my own.
It'll take time for me to get back into this project, but it's still very functional. Here are some links:
- 25+ example videos (some with code): https://labofoz.notion.site/901ff075d8df4bd7b9500d134c22038b?v=f33d13f1c67b458487250992a8501e2c
- WordPress version of docs: https://handsfree.dev
- Gesture mapper for hands: https://handsfree.dev/tools/gesture-mapper/
- The lightweight "offline" docs (what you get when you run
npm start
): https://handsfree.js.org
Thanks for your patience, and also for all the support I have received over the last 3 years of developing this library! Oz
This repo is broken into 3 main parts: The actual library itself found in /src/
, the documentation for it in /docs/
, and the Handsfree Browser Extension in /extension/
.
- Handsfree.js
- Quickstart with CDN (fastest way to get started)
- Quickstart with NPM (requires server or bundler)
- Development guide
- Handsfree Browser Extension
Note: models loaded from the CDN may load slower on the initial page load, but should load much faster once cached by the browser.
This option is great if you don't have or need a server, or if you're prototyping on a site like CodePen. You can also just download this repo and work with one of the /boilerplate/
.
<head>
<!-- Include Handsfree.js -->
<link rel="stylesheet" href="https://unpkg.com/[email protected]/build/lib/assets/handsfree.css" />
<script src="https://unpkg.com/[email protected]/build/lib/handsfree.js"></script>
</head>
<body>
<!-- Your code must be inside body as it applies classes to it -->
<script>
// Let's use handtracking and show the webcam feed with wireframes
const handsfree = new Handsfree({showDebug: true, hands: true})
handsfree.start()
// Create a plugin named "logger" to show data on every frame
handsfree.use('logger', data => {
console.log(data.hands)
})
</script>
</body>
# From your projects root
npm i handsfree
// Inside your app
import Handsfree from 'handsfree'
// Let's use handtracking and enable the plugins tagged with "browser"
const handsfree = new Handsfree({showDebug: true, hands: true})
handsfree.enablePlugins('browser')
handsfree.start()
The above will load models, some over 10Mb, from the Unpkg CDN. If you'd rather host these yourself (for example, to use offline) then you can eject the models from the npm package into your project's public folder:
# Move the models into your project's public directory
# - change PUBLIC below to where you keep your project's assets
# ON WINDOWS
xcopy /e node_modules\handsfree\build\lib PUBLIC
# EVERYWHERE ELSE
cp -r node_modules/handsfree/build/lib/* PUBLIC
import Handsfree from 'handsfree'
const handsfree = new Handsfree({
hands: true,
// Set this to your where you moved the models into
assetsPath: '/PUBLIC/assets',
})
handsfree.enablePlugins('browser')
handsfree.start()
The following aims to give you a quick overview of how things work. The key takeaway is that everything is centered around hooks/plugins, which are basically named callbacks which are run on every frame and can be toggled on and off.
The following workflow demonstrates how to use all features of Handsfree.js. Check out the Guides and References to dive deeper, and feel free to post on the Google Groups or Discord if you get stuck!
// Let's enable face tracking with the default Face Pointer
const handsfree = new Handsfree({weboji: true})
handsfree.enablePlugins('browser')
// Now let's start things up
handsfree.start()
// Let's create a plugin called "logger"
// - Plugins run on every frame and is how you "plug in" to the main loop
// - "this" context is the plugin itself. In this case, handsfree.plugin.logger
handsfree.use('logger', data => {
console.log(data.weboji.morphs, data.weboji.rotation, data.weboji.pointer, data, this)
})
// Let's switch to hand tracking now. To demonstrate that you can do this live,
// let's create a plugin that switches to hand tracking when both eyebrows go up
handsfree.use('handTrackingSwitcher', {weboji} => {
if (weboji.state.browsUp) {
// Disable this plugin
// Same as handsfree.plugin.handTrackingSwitcher.disable()
this.disable()
// Turn off face tracking and enable hand tracking
handsfree.update({
weboji: false,
hands: true
})
}
})
// You can enable and disable any combination of models and plugins
handsfree.update({
// Disable weboji which is currently running
weboji: false,
// Start the pose model
pose: true,
// This is also how you configure (or pre-configure) a bunch of plugins at once
plugin: {
fingerPointer: {enabled: false},
faceScroll: {
vertScroll: {
scrollSpeed: 0.01
}
}
}
})
// Disable all plugins
handsfree.disablePlugins()
// Enable only the plugins for making music (not actually implemented yet)
handsfree.enablePlugins('music')
// Overwrite our logger to display the original model APIs
handsfree.plugin.logger.onFrame = (data) => {
console.log(handsfree.model.pose?.api, handsfree.model.weboji?.api, handsfree.model.pose?.api)
}
Face Pointers |
Motion Parallax Display |
Puppeteering Industrial Robots |
Playing desktop games with face clicks |
Hand Pointers |
Use with Three.js |
Playing desktop games with pinch clicks |
Laser pointers but with your finger |
Flappy Pose - Flappy Bird but where you have to flap your arms |
If you'd like to contribute to the library or documentation then the following will get you going:
- Install NodeJS and git
- Clone this repository:
git clone https://github.com/handsfreejs/handsfree
- Install dependencies by running
npm i
in a terminal from the project's root - Start development on
localhost:8080
by runningnpm start
- Hit CTRL+C from the terminal to close the server
Once you've run the above, you can just use npm start
. If you pull the latest code, remember to run npm i
to get any new dependencies (this shouldn't happen often).
# Start local development on localhost:8080
npm start
# Builds the library, documentation, and extension
npm run build
# Build only the library /dist/lib/
npm run build:lib
# Build only the documentation at /dist/docs/
npm run build:docs
# Build only the extension at /dist/extension
npm run build:extension
# Publish library to NPM
npm login
npm publish
# Deploy documentation to handsfree.js.org
deploy.sh
- See vuepress-component-font-awesome for adding new icons to the documentation. Remember to run
npm run fa:build
when adding new font icons so that they are copied over into thedocs/.vuepress/components/FA
folder - You may occasionally need to restart server when adding new files to the
/docs
, this is true when changing/docs/.vuepress.config.js
as well
The Browser Extension is a designed to help you browse the web handsfree through face and/or hand gestures. The goal is to develop a "Userscript Manager" like Tampermonkey, but for handsfree-ifying web pages, games, apps, WebXR and really any other type of content found the web.
- When you first install the extension,
/src/background/handsfree.js
checks if you've approved the webcam. If not, then it'll open the options page insrc/options/stream-capture.html
- The popup panel has a "Start/Stop Webcam" button that communicates with the background script to start the webcam:
/src/popup/index.html
- The background page is where the models are stored and run. This keeps everything isolated and only asks for webcam permission once (vs on every domain):
/src/background/handsfree.js
- The background page also uses the "Picture in Picture" API to "pop the webcam" out of the browser. It renders the webcam feed and debug canvases into a single canvas, and uses that as the
srcObject
to a separate video element which is the PiP'ed
Install as an unpacked chrome extension.
- Visit
chrome://extensions
- Enable Developer Mode on the top right
- Then click Load unpacked and select this project's root folder
By default, each page will get a "Face Pointer" or a set of "Palm Pointers" for you to browse pages handsfree.
However, in addition to the pointers you can add custom handsfree interactions. For example, here's a demonstration of handsfree-ifying different things:
Explore the interactive docs at: Handsfree.js.org
Or try it right away with the serverless boilerplates in /boilerplate/
!
The Handsfree.js core is available for free and commercial use under Apache 2.0. Each of the models are also available for free and commercial use under Apache 2.0:
- Jeeliz Weboji (Apache 2.0)
- MediaPipe Models (Apache 2.0)
- TensorFlow.js Models (Apache 2.0)
I'd like to also thank the following people and projects:
- 98.css by @jdan (MIT) - Used as boilerplate for documentation theme
- handpose-facemesh-demos by [@LingDong-] - Used as a boilerplate for the Handpose Three.js setup
- @Golan and the The STUDIO for Creative Inquiry for hosting me for a residency during 2019 and for helping me approach projects in a more expressive way. Also for inviting me back for a multi-month residency in Spring 2021!
- @AnilDash for supporting the project during Winter 2018 out of the blue and the opportunities to share my project on Glitch.com
- The School of AI for the 2018 Fellowship in support of this project
- @jessscon and Google PAIR for the very early support that made starting this project possible
- Everyone who's previously supported the project through GitHub Sponsors, Patreon, GoFundMe, and through Twitter and everywhere else over the years