Real-time
Expressive avatars on everyday CPUs
Lip-synced avatars on Mac mini, Raspberry Pi, and commodity hardware. No GPU. Self-hostable.
- 25 FPSreal-time output
- Any CPUno GPU required
- 1.1sto first frame
- ~178MBvisually lossless bundle
Runs on the hardware you have
From a Raspberry Pi to a laptop CPU — no dedicated GPU, no cloud bill.
Lower cost at scale
No GPU infrastructure across kiosk fleets and embedded devices.
Privacy by default
Compute on-device — works with limited or no connectivity.
One portable bundle
A single HEVC file you drop on any device — no runtime downloads.
Quickstart
Grab an API secret, add a .imx model, render frames.
# pip install bithuman
import asyncio
from bithuman import Bithuman
async def main():
bh = Bithuman(api_secret="YOUR_API_SECRET", model="agent.imx") # from dashboard
async for frame in bh.speak("Hello, I'm an Essence avatar."):
# 25-FPS RGB frame (no audio output by default — pair with your TTS pipeline)
handle(frame)
asyncio.run(main())Technical details
How it stays fast on CPU
Re-engineered CPU-native: reuses pre-computed patterns to cut the redundant compute of a naive port by up to ~3×. Bundle is ~178MB (HEVC), no visible quality loss.
Where it runs
- Mac mini and compact desktops
- Raspberry Pi-class embedded devices
- Commodity hardware for scalable rollouts
Best fit
- Retail and hospitality kiosks
- Museums and interactive exhibits
- On-prem enterprise assistants
- Field deployments with limited connectivity
Build with bitHuman Essence
Ship your first character in minutes — from local prototype to production at scale.