Open-LLM-VTuber: Talk to LLMs with a Live2D Avatar

Overview

Open-LLM-VTuber connects voice conversation, language models, and a Live2D character in an AI companion that can run on Windows, macOS, and Linux. It addresses the gap between text-based chat interfaces and hands-free, character-driven interaction, with both web and desktop usage modes.

It is suited to people who want to customize an on-screen companion and choose among local or hosted LLM, speech-recognition, and speech-synthesis options. The project emphasizes local operation, though using cloud APIs is also supported.

Key Features

• Voice conversations with interruption support, including a mode designed to avoid the assistant hearing its own speech.
• Live2D avatars with configurable character appearance, persona, and expression mapping.
• Support for local and hosted LLMs, including Ollama and OpenAI-compatible APIs.
• Multiple speech recognition and text-to-speech backends, including offline options.
• Web and desktop clients, with a transparent desktop-pet mode in the desktop client.
• Optional visual input from a camera, screen recording, or screenshots.
• Persistent chat logs for returning to earlier conversations.
• Modular configuration for switching backends and extending integrations.

Use Cases

• A hobbyist can set up a customized voice-driven character for casual conversation or companionship.
• A Live2D creator can connect an avatar to an LLM and configure its persona and expressions.
• A privacy-conscious user can run compatible local models and speech components on their own computer.
• A streamer or desktop user can keep an interactive avatar visible in a desktop-pet window while working or chatting.

Getting Started

Follow the Quick Start documentation for installation and setup. Configuration varies by platform and by the selected model and speech backends; see the repository README for project details.

Alternatives

• QwenPaw: QwenPaw is a general-purpose personal assistant for chat applications, while Open-LLM-VTuber centers on hands-free voice conversation with a Live2D avatar.

Considerations

The README describes the project as under active development. The v2.0 rewrite is in early discussion and planning, and maintainers ask contributors to direct v1 feature requests to the developer community rather than opening new issues or pull requests. Existing v1 bugs and pull requests are still being addressed.

The project supports a range of backends, but setup and resource needs depend on the selected models and services. Remote browser access that uses the microphone requires HTTPS or localhost. The repository reports its license as NOASSERTION, and bundled Live2D sample models are covered by separate Live2D terms, not the project license.