Afterglow
A live-streaming setup for one person, with a real-time 3D avatar on camera instead of the streamer.
Overview
Afterglow replaces the streamer on camera with a real-time 3D avatar. One person runs the whole stream, with the kind of production that usually needs a crew.
Hands one to one
The avatar's fingers have the same lengths as the streamer's, so no fitting is needed.
LED face
An LED-matrix face shows emotions read from the streamer's face.
Two voice paths
The stream gets the voice with a mood mask. The game gets the clean voice.
Supervised services
A supervisor restarts any service that stops. The streamer controls everything from a phone.
Architecture
Picture, sound and viewer input run in parallel. They meet in the compositor, which combines the scene, the game and the overlays into the broadcast.
Anatomy of the rig
The same system, drawn as a room. The desk and cameras are in front, the machines are on the right, and behind them is the stage that the scene renders. Hover over a part to read what it does.
One to one
Most avatar setups use a stock hand and a solver that bends it until it looks close. Afterglow sets the avatar's finger bones to the lengths of the streamer's fingers, so the joint angles and fingertips match without fitting. The engine reports the final joint positions back, which makes the error measurable.
Services
Small services run around the scene. The supervisor restarts any of them that stops.
| Service | What it does |
|---|---|
| Overlays | Graphics over the stream, including the chat line |
| Music | Tracks, including the ones viewers order |
| Video wall | The giant LED wall behind the avatar |
| LED face | The avatar's emotions, from the face tracking |
| Dashboard | Controls the stream from a phone |
| Viewer site | Shows the stream in a CRT style. Viewers order tracks and spend points there |
Where it stands
Live on stream. The full stack runs today, and measuring hand accuracy comes next.
Works today
- The full stream stack
- The avatar scene
- The voice chain
- The LED face and the video wall
- Overlays, viewer site and dashboard
Next
- Measured verification of hand accuracy
- Three-camera body tracking
- Finer lighting