Built for global and cross-border clients — USDT needs no cards, settles fast and works across borders. Large deals can arrange other methods with support.

Do you support private deployment?

Yes. Chat aggregation and self-controlled AI can both deploy to your own servers — data stays local with no cloud reporting.

How is your AI translation different from Google Translate?

We use AI translation + chat tech that outputs native slang, local idioms and cultural tone — they can't tell you're foreign — unlike tools that wire up Google-style APIs and read stiff and literal.

Can AI close deals automatically? Can humans take over?

Yes. AI works your persona 24/7 to engage, answer, follow up and convert; at key moments a human can take over in one click.

What input do voice cloning / face swap need?

Voice cloning needs a short clear voice sample; face swap needs clear face images or video. Ensure you hold the rights.

How is a private LLM different from a public cloud API?

A privately deployed LLM keeps data fully local, free of public-cloud dependencies — fine-tune its knowledge base and output style to your business, with no cloud reporting and full self-control.

Yes. Most services offer both subscriptions and usage-based add-ons — pay for what you use, mix freely.

How do I avoid payment scams?

Always verify the latest receiving address via our official Telegram support before paying. Never trust addresses forwarded by third parties.

For v1.0.11 · continuously updated

AvatarHub User Manual

The complete guide from install to going live: requirements, activation, the four core workflows, updates and troubleshooting.

Download the client

01Product overview

AvatarHub is a locally deployed real-time digital human engine: voice cloning, live face swap, digital-human streaming and cloned-voice interpreting run on your own PC or server — data never leaves your premises. Core pipeline: speech → STT → LLM → cloned-voice TTS → real-time lip sync / face swap → live stream or video call.

Voice cloning: zero-shot cloning from a 10-second sample, realistic multi-language synthesis.
Live face swap: real-time on-camera presence for streams / video calls, OBS virtual camera supported.
Digital-human streaming: cloned voice + real-time lip sync with interactive Q&A.
Cloned-voice interpreting: speak Chinese, your audience hears a foreign language in your own voice.

This manual is for users of the official installer. For multi-machine clusters or custom development, contact support for the deployment guide.

02System requirements

Item	Minimum	Recommended
OS	Windows 10 (x64)	Windows 11 (x64)
GPU	NVIDIA 4 GB VRAM (voice cloning only)	NVIDIA 8 GB+ (live swap / digital human); 24 GB (full interpreting suite)
RAM	16 GB	32 GB
Disk	40 GB free on SSD	80 GB free on SSD
Network	Internet required for first-time component download	50 Mbps+ for faster downloads

The macOS 12+ (Apple Silicon / Intel) build is a lightweight console / remote client: heavy inference (face swap, digital human) still runs on a Windows / server NVIDIA GPU, with the Mac connecting remotely. Coming soon.

Not sure what your machine can run? The first-run wizard benchmarks your hardware and tells you exactly which features are available — or ask the AI assistant in the corner.

03Download & install

1
Download the installer (~1 min)
Click the download button on this page to get AvatarHub-Setup-x.x.x.exe (a ~45 MB ; AI components download on demand later). Optionally verify the shown on this page: run certutil -hashfile <installer path> SHA256 in PowerShell and compare.
2
Run the installer (~1 min)
Double-click the installer and follow the wizard. It installs per-user with no admin rights and lets you choose the folder. An AvatarHub icon appears on the desktop and Start menu when done. Common snag: Most common snag: the "Windows protected your PC" prompt — a routine notice for recently published apps, not a virus alert. Click More info → Run anyway; verify the first if you like.
3
First launch · hardware detection (~2 min)
Open AvatarHub. The detects your GPU and recommends a (Lite / Standard / Flagship) — zero command line. Lite (4–6 GB [[VRAM]]): voice cloning, photo / video face swap. Standard (8 GB+ VRAM, RTX 3060 and up): live face swap, digital-human streaming. Flagship (24 GB VRAM, RTX 4090 / 5090): real-time HD plus the full interpreting suite.
4
Automatic component download (~10–30 min (bandwidth))
Confirm the tier and the wizard downloads the required AI components and models — every file verified, all downloads . Expect 10–60 GB depending on tier; keep 80 GB free on an SSD. Feel free to explore the UI meanwhile. Common snag: Slow or interrupted? Just reopen AvatarHub — continues from the breakpoint and finished parts are never re-downloaded. On weak networks, switch to a steadier connection.
5
Activate or start the trial (~1 min)
Go to Settings → License, enter your order number for (the signed license is fetched automatically) — or just start the 14-day free trial, no sign-up required.
6
Verify the install (~5–10 min)
Click Start All in the launcher and wait for the service lights to turn green; run the health check — all green means you're done. Then a quick : upload a clear 10-second voice sample in the console → clone → play. If you hear it, everything works.

Stuck during install? See the Troubleshooting chapter, or click AI install assistant on the download page and let the AI walk you through it.

04Activation & trial

14-day free trial: starts automatically after install — no steps needed.
Online activation: after purchase, enter your order number in Settings → License; the signed license is fetched and applied instantly.
Licenses are hardware-bound: contact support before moving to a new machine.

Check license status any time in Settings → License: current version, license type and expiry.

05Quick start · clone a voice and talk

1
Upload a voice sample
Console → Characters/Voices → upload a clear ~10-second voice clip (minimal noise, not too short).
2
Clone and preview
Click Clone → Preview; happy with it? Activate the character.
3
Start talking
Open the chat page (PC) or the phone page (same-WiFi phone), type or hold the mic — the digital human replies in the cloned voice with lip sync, and you can barge in mid-sentence.

The first mic press on mobile asks for permission — tap Allow. If you tapped Block, use the address-bar 🔒 → Site settings → Microphone → Allow, then refresh.

06Live face swap for streams / video calls

1
Install virtual devices
Install OBS Studio () and VB-Cable () — one-time setup.
2
Prepare an on-camera character
Create / activate a character in the console (upload the target face).
3
Device check, then go live
On the streaming page run the Device Check; once mic / camera / virtual audio are all green, click Go Live.
4
Select virtual devices in the target app
In WeChat / TikTok / Zoom, set the camera to OBS Virtual Camera and the microphone to CABLE Output.

For portrait video calls choose the 720×1280 portrait aspect. No physical camera? Use your phone as the camera (next chapter).

07Phone interpreting (speak Chinese, they hear your language)

1
Start interpreting services on the PC
Click Start All in the launcher and confirm the live-interpreting service is running.
2
Scan the QR code
The PC shows two QR codes — scan ① Interpreting. Phone and PC must be on the same WiFi.
3
Trust the local certificate
The "connection is not private" warning is expected (self-signed local cert): tap Advanced / Details → Proceed.
4
One-tap setup
Tap 🚀 Prepare → allow the microphone; when the guide bar turns green, start speaking. Tap 📷 to add camera if you need to be on screen.

The guide bar at the top of the phone page tells you the next step at every stage, including how to recover if you tapped Block. High latency? Switch to 5 GHz WiFi or enable low-latency mode.

08Updates & rollback

In-app self-update: when a new version is available the launcher shows v-current → v-new — one click downloads, installs and restarts in about 1–3 minutes.
Your data stays: updates replace only the console program (tens of MB); downloaded AI components and characters are untouched.
Verified updates: release manifests carry an ; tampered packages are rejected.
One-click rollback: not happy with a version? Roll back to the previous one from Software Update — data preserved.
Stream-safe: updates never apply while a live stream / interpreting session is running.

Per-version changes are listed in the Release notes section of the download page.

09Troubleshooting

Symptom	Fix
Console won't open	Run the launcher health check; restart services; hard-refresh with Ctrl+F5
A service stays red / not ready	Usually low VRAM or models still loading: wait or restart it; disable unused extensions
Out of GPU memory (OOM)	Run core features only (disable extras); HD lip sync is the heaviest — step down a tier
Phone can't connect	Phone and PC must share the same WiFi; allow the firewall prompt
"Connection is not private"	Expected (self-signed local cert): Advanced → Proceed
Microphone won't enable	Address-bar 🔒 → Site settings → Microphone → Allow → refresh
No interpreting subtitles	Make sure the live-interpreting service is running (green in the launcher)
Face swap shows the original face	Confirm a character is activated and the face-swap service is running

Still stuck? Use the (~3–10 s, auto-uploaded) and give support the 6-digit code — they can pinpoint the issue remotely, no manual file sending.

10Glossary

Terms with a dotted underline in the guides show an explanation on hover or tap; this chapter collects them all for printing and quick reference.

Term	Meaning
thin-core installer	The installer itself is only ~45 MB; the multi-GB AI components and models download on demand at first launch instead of shipping in one huge package.
SHA-256	A unique digital fingerprint of a file. If the checksum you compute matches the one published here, the installer is intact and untampered.
SmartScreen	Windows' built-in reputation check. Recently published installers routinely trigger the "Windows protected your PC" prompt — it is not a virus alert. Verify the SHA-256, then click More info → Run anyway.
first-run wizard	The guided flow on first launch: detects your hardware, recommends a capability tier and downloads the required components — no command line involved.
capability tier	Feature levels (Lite / Standard / Flagship) mapped to your GPU. The tier decides which AI features are available and how much gets downloaded. Check any time in Settings → Hardware tier.
VRAM	The GPU's dedicated memory. AI models load into VRAM to run — more VRAM means more features can run simultaneously at higher quality tiers.
resumable download	Interrupted downloads continue from where they stopped — finished parts are never re-downloaded, even across restarts.
online activation	Enter your order number in Settings → License and the client fetches and applies the signed license automatically — no pasting long activation strings.
smoke test	The quickest end-to-end check: clone a 10-second voice sample and play it back. If you hear it, the whole STT → synthesis → playback pipeline works.
OBS Virtual Camera	A virtual webcam provided by OBS Studio: the AI-processed video shows up as a camera device that WeChat / TikTok / Zoom can select directly.
VB-Cable	A virtual audio driver: AI-processed audio becomes a microphone device (CABLE Output) that chat apps can pick as their mic input.
Ed25519 signature	A cryptographic signature on release manifests. The client only accepts updates signed by the official key; tampered packages are rejected outright.
diagnostic pack	A built-in troubleshooter: collects redacted logs in ~3–10 s, uploads them and returns a 6-digit code. Give support the code — no manual file sending.

11Getting help

AI assistant: the chat bubble in the corner, 24/7 instant answers on installing, usage and pricing.
Human support on Telegram: @WJKJ2026 (about 5-minute response during working hours).
Remote installation: 99 USDT — book a session and we set everything up.
Diagnostic code: generate a 6-digit code with the in-app and share it with support.

Not covered here? Ask the AI assistant in the corner, or contact @WJKJ2026.