danmuai:基于 Python 的 Windows 桌面 AI 弹幕工具项目

可帮助直播主播生成并展示AI弹幕,提升直播互动体验。该工具截取显示器全屏,调用视觉模型生成5条弹幕,通过Qt透明置顶浮层滚动展示,支持Web控制台配置与启停,注重轻量与隐私保护。【此简介由AI生成】

Branch1Tags2

DanmuAI (English summary)

Python License

Windows desktop overlay danmaku (scrolling comments) powered by vision models. Captures the selected display, sends compressed in-memory screenshots to your configured API, and renders transparent Qt overlays. Default UI is a local Web console (127.0.0.1:18765) in a pywebview shell.

Full documentation is in README.md (Chinese). Agent/contributor notes: AGENTS.md.

Quick start

pip install -r requirements.txt
python main.py

Platform

  • Windows with WebView2 (pywebview)
  • Python ≥ 3.12
  • Config and secrets: %APPDATA%/DanmuAI/ (see SECURITY.md)

API providers (Assistant settings)

Built-in presets (see app/model_providers.py) include Volcengine Ark (Doubao), Alibaba DashScope, Zhipu, Moonshot, SiliconFlow, Xiaomi MiMo, plus custom OpenAI-compatible and Doubao Responses entries.

Preset Default endpoint Protocol Vision danmu example
Volcengine Ark https://ark.cn-beijing.volces.com/api/v3 Doubao /responses doubao-seed-1-6-flash-250828
Xiaomi MiMo https://api.xiaomimimo.com/v1 OpenAI-compatible mimo-v2.5 (recommended), mimo-v2-omni

Platforms with a model picker catalog: Doubao, DashScope, SiliconFlow, MiMo (GET /api/model-catalog).

Thinking mode is always sent as disabled on outbound requests (app/ai_client.py). Streaming parsers use content only, not reasoning_content.

Microphone mode sends input_audio only on the Doubao Responses path. OpenAI-compatible presets (including MiMo) are screenshot + text only; use a Doubao multimodal model such as doubao-seed-2-0-mini-260428 for mic.

Contributing

License

SPDX-License-Identifier: GPL-3.0-or-later — see LICENSE and THIRD_PARTY_NOTICES.md.

Introduction

可帮助直播主播生成并展示AI弹幕,提升直播互动体验。该工具截取显示器全屏,调用视觉模型生成5条弹幕,通过Qt透明置顶浮层滚动展示,支持Web控制台配置与启停,注重轻量与隐私保护。【此简介由AI生成】

Customize your domain