Why the face is rendered on your Mac and not streamed from a server
Streaming a face from the cloud costs money every minute and puts a server between you and the conversation. Facelet draws its character on your Mac’s graphics chip instead. The trade-offs, plainly.
There are two ways to put a live face on a screen. Render it on a server and stream video to the browser, or render it on the computer in front of the person. Nearly every talking avatar you can use today does the first. Facelet does the second, and we want to be clear about why, including what it costs you.
What streaming buys, and what it charges
Streaming works on anything with a browser, which is why the business platforms use it: their customers’ customers could be on any device. The bill for that is a server rendering video for every minute of every conversation, passed on per minute, and a face that stutters when the network does. It also means the service is in the middle of the conversation, because the audio has to reach the renderer.
What rendering locally buys
Facelet draws its character on your Mac’s graphics chip, frame by frame, and analyses the voice on your Mac to move the mouth and face. Three things follow. The frame rate is your Mac’s, not the network’s. There is no per-minute cost, so there is no per-minute price. And the only parties to the conversation are you and OpenAI, because you are talking to ChatGPT on your own account; nothing about it passes through Facelet’s servers. What we learn about you is your email, your purchase and the Mac you activated on.
What it charges
Rendering a lifelike human is real work, so Facelet needs an Apple silicon Mac with an M2 chip or newer, macOS 14 or newer, and 16 GB of memory is recommended. Intel and M1 Macs are not supported by the current build. There is no iPhone, iPad, Windows or browser version, and we are not planning a streamed one, because it would undo the three things above. Development has been on an M2 Ultra, and we are still confirming performance across the supported range before the beta ships.
The short version
A streamed face is a service. A rendered face is a thing you own. Facelet is the second, for one payment, and the details are on the rendered on your Mac page and in the privacy policy.