Make a face, and Dash makes it back.

The camera plugin for web finds the user’s face in the frame. The TensorFlow.js MediaPipe FaceMesh model then maps facial features live, inside the camera frame.

The result is a futuristic photo booth where your expressions show up in real time on a virtual Dash or Sparky.

To animate the background and the avatars, we turned to Rive, a web app built in Flutter for high-performance animation. Rive state machines control how Dash and Sparky respond to movement, so the avatar mimics the user.

FaceMesh’s feature detection is matched to coordinates on the avatar models, and a Rive state machine controller turns that input into what the avatar looks like on screen.

With Cloud Functions for Firebase and Cloud Storage for Firebase, Holobooth turns a session into a GIF or video that users share to social media with one click.

Friends who open the post can step into the booth and make their own.

We generated the project with Very Good CLI, so the first commit already had null safety, internationalization, and 100% unit and widget test coverage.

The code is split into a layered architecture, and all of it is open source in the Flutter repository on GitHub.