Digital human
Mirnak
A photoreal digital human that listens, thinks, and answers out loud. Speech to lipsync in under two seconds, streamed to any browser.
The loop
Listen
Speech captured in the browser and transcribed.
Think
A language model reads it and decides what to say.
Speak
A neural voice turns the answer into audio.
Perform
The audio drives real time lipsync on a photoreal face.
Stream
The rendered face is streamed back as video.
How it works
The hard part is not the face
Anyone can render a head. The engineering is in holding the conversational loop fast enough that a person forgets they are talking to software.
The whole loop, under two seconds
Speech in, transcription, language model, neural voice, facial animation, and video back out. Every hop costs time, and past a couple of seconds the illusion collapses and you know you are talking to a machine. The whole round trip is held inside two seconds.
A real face, not an avatar cartoon
A photoreal MetaHuman rendered in Unreal Engine 5, with facial animation driven from the audio in real time. Rendered on a cloud GPU and streamed, so the browser does no work.
It runs in a tab
No install, no plugin, no download. Pixel streaming means the heavy rendering happens on a GPU somewhere else and the user gets video and a microphone.
Multilingual, including dialect
It can take speech in a regional dialect and answer in the standard written form of the language, which is what most real deployments actually need.
Talk to it yourself
The demo is live in your browser. Give it a microphone and have a conversation.