A Windows desktop AI danmaku assistant that periodically captures the screen, has a vision model generate danmaku and shows them on a transparent overlay, and can also read danmaku aloud via TTS, providing an observing and commenting AI audience for streams, recordings, games and desktop content.
- v0.5.0
- Published at
- 15d ago
- 360
- Stars
- 1,924
- Download
- Synced at
- 14d ago
Latest release
v0.5.0Oct 4, 2026, 02:32 PMDownloads slow? Switch the source anytime above
README
Project Overview
DanmuAI is a desktop AI danmaku assistant for Windows. It captures the screen of a chosen monitor at a set interval, calls a vision model to understand the current content and generate danmaku, then shows them on a transparent always-on-top overlay. A local web console lets you configure the model, persona, monitors, danmaku style and runtime options.
Give streams, recordings, games and desktop content an AI audience that watches and reacts
Key Features
- Visual danmaku: Captures the screen on a timer and lets a vision model generate danmaku related to the current content
- Custom persona and model: Configure persona, model platform, endpoint and custom models to shape the tone
- Transparent overlay: Scrolls danmaku on a chosen monitor, independent of whether the web console is open
- AI reading: Supports MiMo, Alibaba Bailian Qwen3, MiniMax and Doubao V3, with a voice catalog, caching, refresh and preview
- Adjustable appearance: Supports floating panel animation, built-in styles and managed CSS themes
- Multi-monitor support: Choose separate monitors for capture and for the danmaku overlay
- Optional modules: Include microphone input, VTuber Live2D model import, meme danmaku and a knowledge base
Target Audience
- Streamers: Hosts who want AI danmaku for live or recorded content
- Content viewers: Users who want an AI audience to comment on games or desktop content
- TTS users: Users who need AI voice reading of danmaku
- Advanced users: Users willing to configure a model platform and API key themselves
Quick Start
- Download the latest Windows version from the Releases page: Setup.exe for install, Portable.zip for a portable copy
- After launch, fill in the model platform, API key and model info in the web console
- Choose the capture monitor, the overlay monitor, the recognition interval, danmaku count per batch, persona and style
- Start it, and danmaku appear on the chosen monitor
- For voice reading, pick the provider, voice, reading interval and style on the AI Reading page
Core Advantages
- Sees the screen: Uses screenshots and a vision model so danmaku match what is on screen
- Local first: The console listens only on the local machine by default, and screenshots are compressed in memory rather than written to disk
- Freely configurable: Model, persona, monitors and danmaku style can all be adjusted
- Does not interrupt viewing: The overlay runs independently of the console, so closing the console does not stop display
System Requirements
| Item | Requirement |
|---|---|
| OS | Windows |
| Architecture | Windows desktop |
| Runtime | Python 3.12 to run from source; WebView2 for the default desktop shell |
Notes
- Credential safety: Put the API key only in the app configuration, never in a repository or public screenshot
- Local storage: Configuration, model info and credentials are stored locally under %APPDATA%/DanmuAI/
- Data egress: Whether images or text are sent to an external service depends on the model platform you configure
- Responsibility: Use of AI-generated content and third-party services is the user's own responsibility
Release history
Sort