FaceStream

This is a web app that let's you swap your face in real-time without having a powerful machine. You can try it out live here.

See the gif below for a demo:

Motivation

I discovered the Deep Live Cam project and tried to run it on my M1 MacBook, but only got 0.5 FPS. So I wanted to explore how fast and at what latency you could get this to run on a remote server via WebRTC.

How to run this yourself

Prerequisites

To run this yourself you wil need:

• uv
• A modal account as well configured credentials

Then you can run the following command to start the server:

uv run modal serve facestream.main

Or this command to deploy it:

uv run modal deploy -m facestream.main

Optional: TURN Server for Cellular Networks

On most cellular networks you need a TURN server for WebRTC to work. You can create a TURN app on Cloudflare. To use those with this app you need to:

1. Create a secret called facestream in modal with the following values:

TURN_TOKEN_ID=your-turn-token-id
TURN_API_TOKEN=your-turn-api-token

2. Comment out the following line in src/facestream/main.py:

...
secrets=[
    modal.Secret.from_name(
        "facestream",
        required_keys=[SECRET_KEY_TURN_TOKEN_ID, SECRET_KEY_TURN_API_TOKEN],
    )
]
...

Credits

• This project was inspired by and uses the model of Deep Live Cam. Go check out their project. They have some extra features that I didn't implement here. Note that the model used in that project (and therefore also this one) is only for non-commercial use.

• Modal made this easy to build and deploy. They are generously providing free credits to host the live demo.