Most talking avatars are a video stream
The usual way to put a live face on the screen is to render it on a server and stream the video to a browser tab. It works anywhere, and it costs the company money every minute you are connected, which is why it is billed that way. It also means the face buffers when the network hiccups, and a server sits in the middle of everything you say.
Facelet draws the face on your Mac
Facelet is a native macOS app. The character is rendered by your Mac’s graphics chip, frame by frame, while you talk. The audio Facelet analyses to move the mouth and face is analysed on your Mac and never uploaded. There is no video stream, no buffering, and no rendering server. Close the lid and it stops; open it and it is there, without a queue.
What leaves your Mac, and what does not
| What | Where it goes |
|---|---|
| Your voice, and ChatGPT’s answers | Between your Mac and OpenAI, on your own account, under OpenAI’s terms |
| The face and its animation | Rendered on your Mac; nothing uploaded |
| The audio analysed for lip and face movement | Analysed on your Mac; nothing uploaded |
| Your ChatGPT password, history and memories | Never seen by Facelet |
| Your email, purchase and activated Mac | Facelet, to sell you a seat and issue a license |
The full account is in the privacy policy, written to be read, and in what stays on your Mac.
Why Apple silicon
A lifelike human at a steady frame rate is a real workload: skin, hair, eyes, and a mouth being driven by live audio, all while a voice conversation runs alongside. Apple silicon’s graphics and unified memory handle it on a laptop without a fan roar. Facelet needs an M2, M3, M4 or newer chip, macOS 14 or newer, and 16 GB of memory is recommended. Intel and M1-family Macs are not supported by the current build, and there is no iPhone, iPad or Windows version. Development has been on an M2 Ultra; performance across the supported range is still being confirmed.
What you gain
- No per-minute cost. Rendering on your machine costs us nothing, so we do not charge for it. See one payment, no API key.
- A face that does not buffer. The frame rate depends on your Mac, not on the network.
- A shorter list of who hears you. You and OpenAI, because you are talking to ChatGPT. Not us.
- An app you own. A license on your Mac, not a session on someone’s server that can be switched off.
Questions
Does Facelet work offline?
The face does; the conversation does not. Rendering happens entirely on your Mac, but talking requires ChatGPT, so you need an internet connection to your ChatGPT account. Nothing else is required.
Why Apple silicon only?
Rendering a lifelike human at a smooth frame rate while a voice conversation runs takes a modern graphics chip and unified memory. Facelet needs an M2 or newer chip and macOS 14 or newer, with 16 GB of memory recommended. Intel and M1 Macs are not supported by the current build.
What does Facelet upload?
Your voice goes to OpenAI, because you are talking to ChatGPT through your own account. The audio Facelet analyses to move the mouth is analysed on your Mac and never uploaded. Facelet, the company, learns your email address, your purchase and the Mac you activated on, and nothing about what you say.