bitHuman for developers
bitHuman is a real-time avatar SDK: send it the speech your companion says and it returns a lip-synced face — a photoreal person with Essence 2 or any character with Expression 2 — rendered on iPhone, iPad, Android, Mac, Linux or in the browser.
You bring the persona, the voice and the language model; when the avatar renders in your app on the device and you use your own voice and language services, bitHuman receives usage metering only, never audio, video or conversation text.
Pick a platform below. Each has a copy-paste quickstart in the docs and a working example you can clone.
Deploying in a bank, a hospital or on a show floor? Enterprise →
Platforms
A Swift package that renders the avatar on the iPhone or iPad.
.package(url: "https://github.com/bithuman-product/homebrew-bithuman.git", from: "2.19.0")Two Maven Central artifacts that render the avatar on arm64 phones.
implementation("ai.bithuman:expression2-android:0.5.2") // any character
implementation("ai.bithuman:essence2-android:0.8.1") // a photoreal personThe Swift package for Mac apps, or the CLI for a desktop companion.
.package(url: "https://github.com/bithuman-product/homebrew-bithuman.git", from: "2.19.0")An iframe or a floating widget, in any site or React app.
<iframe src="https://www.bithuman.ai/embed/A23WJF0199" allow="microphone *" style="width:100%;height:600px;border:0"></iframe>Load an avatar, push audio and get frames on macOS or Linux.
pip install "bithuman[expression-2]"There is no native Windows SDK. On Windows, the Python SDK and the CLI run under WSL2, and the web embed runs in the browser.
Architecture
The avatar renders
16 kHz mono speech in, lip-synced frames out
The conversation runs
Performance
| Device | Essence 2 | Expression 2 |
|---|---|---|
| macOS · CLIApple M4 | 4.24× real time2026-09-27 | 8.4× real time2026-09-27 |
| macOS · Swift packageApple M4 | 4.8× real time2026-09-24 | 8.85× real time2026-09-24 |
| iPhone · Swift packageiPhone 15 | 2.16× real time2026-09-27 | 5.55× real time2026-09-27 |
| AndroidSamsung Galaxy S25+ | 2.08× real time2026-09-25 | 2.4× real time2026-09-23 |
| Web browser (WebGPU)Chrome on Apple M4 | 1.72× real time2026-09-27 | 1.95× real time2026-09-27 |
| iPhone · Swift package · held 10 miniPhone 15 | 1.32× real time2026-09-25 | 5.15× real time2026-09-25 |
| Android · held 10 minSamsung Galaxy S25+ | 1.48× real time2026-09-27 | 2.2× real time2026-09-27 |
| Web browser (WebGPU) · held 10 minChrome on Apple M4 | 2.16× real time2026-09-27 | 2.05× real time2026-09-25 |
Pricing
| Item | On the device (Essence 2, Expression 2) | bitHuman cloud avatar | Managed voice chat (all-inclusive) |
|---|---|---|---|
| Credits per minute of active session time | 2 | 4 | 10 |
| At top-up rates ($1 = 100 credits) | about $0.02 | about $0.04 | about $0.10 |
| Minutes in a Creator month (1,800 credits) | 900 | 450 | 180 |
| Speech, language model and voice | You pay your providers | You pay your providers | Included |
| Concurrent sessions | Limited by credits | Per plan (Creator 3 … Enterprise 200) | Per plan (Creator 3 … Enterprise 200) |
Rates from GET https://api.bithuman.ai/v1/pricing, as published on docs.bithuman.ai/pricing.
Privacy
When the avatar renders in your app on the device and you use your own voice and language services, bitHuman receives usage metering only, never audio, video or conversation text. With the web embed, the conversation runs on bitHuman's servers, even when the avatar renders in the tab.
Checklist
The Apple and Android SDKs authenticate with your account's API secret today, so a shipped app carries it. Fetch it from your backend at startup, never compile it in, use a separate secret for each app, and rotate it if it leaks.
What a shipped app holds (docs)Examples
A SwiftUI app with a talking character, rendered on the iPhone or iPad.
One Swift command-line tool: speech in, lip-synced frames out, on your Mac.
Two complete Gradle apps that render a talking avatar on the phone.
android/expression2-hello android/essence2-hello Walkthrough
Open an avatar, play speech through it and watch it talk; conversation.py adds a voice conversation.
A browser front end for an avatar agent in a LiveKit room.
Render a video or run a live avatar from the terminal, with no code.
AI companions
The companion guide walks through a companion screen in an iPhone, iPad or Android app: the avatar renders on the phone and speaks the replies your own voice stack produces — the SDK, the API secret, resampling speech, replies, interruption, and closing the avatar with the screen.
For AI coding agents
FAQ
Yes. The Swift package renders Essence 2 and Expression 2 on iPhone, iPad and Mac. essence2-android and expression2-android on Maven Central render them on arm64 Android phones. Measured results for each device are published at docs.bithuman.ai/performance.
With the Swift and Android SDKs, on the phone. With the web embed, in the bitHuman cloud by default, or in the browser tab with WebGPU (render=local), falling back to cloud rendering. With a LiveKit cloud avatar, in the bitHuman cloud.
The Swift and Android SDKs only render: your app passes in 16 kHz speech and draws the frames. Use any speech recognition, language model and voice, or use bitHuman's managed agent through the web embed. On a Mac or a Linux PC, the bitHuman CLI's local conversation brain can run the whole conversation on the machine.
Yes. Any OpenAI-compatible endpoint works. Set the persona in your model's system prompt or, for a managed agent, in its system_prompt.
Yes. bitHuman provides the companion's face, rendered in real time on the phone, on the Mac or in the browser, and you choose the voice, the language model and the persona. The companion guide in the docs walks through adding the SDK, the API secret, resampling speech, replies, interruption and closing the avatar with the screen.
From 12 October 2026, API and SDK use requires the Creator plan or higher. Rendering on the device uses 2 credits per minute of active session time, talking or idle, which is about $0.02 at top-up rates. A bitHuman cloud avatar uses 4 credits per minute, and the all-inclusive managed voice chat uses 10.
No. Apps on phones and Macs need a connection to start a session, and they keep rendering through a network drop of up to 5 minutes. The web embed needs a connection throughout. Offline license is only available to Business and Enterprise clients who want to run realtime avatars completely locally, off the internet — e.g. kiosks, trade shows, ATM machines, embedded screens. Linux PCs and terminals; arranged through sales.
Usage reports for metering: which session ran and for how long. They never contain audio, video, images or conversation text. The SDK also checks your API secret when a session starts. With the web embed, the conversation itself runs on bitHuman's servers.
Yes, with the iframe embed or the floating website widget. There is no npm package: the embed is an iframe in any framework, and the docs web page has a React snippet.