tegnike/aituber-kit

A web application for chatting with AI characters that anyone can easily set up and deploy.

What is AITuberKit?

AITuberKit is an open‑source toolkit that lets you build a web‑based “AI‑tuber” – a virtual character that can talk, listen, and even stream live on platforms like YouTube. It stitches together:

Large language models (LLMs) – you can plug in OpenAI, Anthropic, Gemini, Ollama, and many other APIs.
Voice synthesis – a long list of TTS engines (VOICEVOX, ElevenLabs, Azure, etc.) turns the model’s text into spoken dialogue.
Avatar rendering – 2‑D (Live2D), 3‑D (VRM) or animated PNG characters are displayed on a web page.
Interaction modes – chat in a browser, respond to YouTube chat, run on a digital‑signage kiosk, or act as a game‑commentary bot.

All of this is packaged as a Next.js application that can be run locally, in Docker, or deployed to serverless platforms such as Vercel or Cloudflare Workers.

Who might use it?

Audience	Why it fits
Content creators / streamers	Turn a virtual avatar into a live host that answers audience comments automatically.
Educators / hobbyists	Quickly prototype an AI‑driven teaching assistant or demo character without writing low‑level code.
Businesses	Deploy a branded AI‑assistant on a website or kiosk, provided a commercial license is obtained.
Developers	Extend the platform via WebSocket APIs, custom prompts, or plug‑in new TTS/LLM services.

How does it work?

Front‑end – A Next.js UI shows the avatar (VRM/Live2D/PNG) and captures microphone or camera input.
Back‑end – Server‑side code forwards user text/audio to the selected LLM or OpenAI Realtime API, receives a response, and optionally calls a TTS service to generate audio.
Memory – Short‑term conversation is kept in‑session; a RAG‑style long‑term store can be configured to pull in past context.
Modes –
- Chat mode – direct browser chat.
- YouTube stream mode – polls YouTube comments (or OneComme) and lets the avatar reply.
- Demo/kiosk mode – full‑screen, password‑protected display with NG‑word filtering.
- Game‑commentary mode – captures screen frames, runs image analysis, and generates live commentary.
Deployment – npm run dev for local dev, Docker Compose for containerised runs, or one‑click Vercel/Cloudflare Workers builds.

What can you do with it today?

Talk to an AI character using any supported LLM.
Add voice with dozens of TTS providers.
Stream on YouTube and let the avatar answer live chat automatically.
Run on a kiosk that greets visitors using face detection.
Enable low‑latency Realtime API for fast back‑and‑forth dialogue.
Hook external apps via WebSocket or a simple HTTP API to send text or images to the avatar.

How easy is it to get started?

git clone … && cd aituber-kit && npm install
Copy .env.example → .env and fill in the API keys you need (OpenAI, ElevenLabs, etc.).
npm run dev (or double‑click LAUNCH.bat / LAUNCH.command).
Open http://localhost:3000 and you’ll see the avatar UI.

Docker users can replace steps 2‑3 with docker compose up -d.

How does it compare to similar projects?

Feature	AITuberKit	Typical “VTuber” bots (e.g., VTube Studio + custom scripts)
All‑in‑one	Provides UI, LLM integration, TTS, streaming, and deployment scripts out of the box.	Usually requires stitching together separate tools.
Model support	VRM, Live2D, animated PNG.	Often limited to one format.
LLM variety	20+ providers, including local Ollama and OpenRouter.	Usually only OpenAI.
Deployment options	Local, Docker, Vercel, Cloudflare Workers.	Mostly local‑only.
Commercial licensing	Custom license for paid use (clear policy).	Varies, often unclear.

Who’s behind it?

The project is a fork of Pixiv’s ChatVRM and is maintained by a community of Japanese‑language developers (see the long list of contributors). The repo is actively starred and forked, and the author provides a demo site, documentation site, and a Discord community for support.

Any red flags?

Custom license – commercial use requires a separate agreement; you must read the license file.
API‑key handling – the backend proxies calls, but you still need to protect your keys in production.
Model assets – the bundled character models have their own restrictive licenses (no commercial redistribution).

TL;DR

AITuberKit is a fairly complete, open‑source platform for creating AI‑driven virtual presenters. It bundles LLM chat, voice synthesis, avatar rendering, and live‑stream integration, and can be run locally or deployed to the cloud. If you want a quick way to turn an AI character into a YouTube‑talking avatar or a kiosk assistant, this toolkit gives you most of the plumbing out of the box, provided you respect the custom licensing terms.