A local Live2D digital human plugin suite for DeepSeek Harness: The Live2D character on the desktop drives lip-sync based on the actual VoxCPM 1.5 audio being played, and switches expressions and actions according to emotion tags.
Stars 26
Reasons
- Lip-sync follows the actual audio envelope sent to the sound card and aligns with measured output latency, rather than estimating based on text length; both speech synthesis and rendering run locally, with no cloud TTS and no account required.
- The desktop pet and TTS integration bridge are two DSH bundles that can be toggled independently; disabling the integration bridge reverts to native TTS. Speech is generated sentence-by-sentence in a streaming fashion, allowing tools to return before playback ends, so the Agent can work while speaking.
Usage
- Requires Windows 10/11 + PowerShell 7, Node 20+, and Python 3.10+; a CUDA-capable GPU is recommended. You must provide your own licensed Live2D Cubism 4 model, VoxCPM 1.5 weights, and a reference voice sample (audio + transcript).
- Run
node scripts/fetch-vendor.mjs --accept-live2d-eula→node scripts/build-all.mjs→ Runlive2d-pet/tools/install.ps1specifying the DSH profile, Python, voice, and model directories, then useverify-install.ps1to check the installation.
Risks
- The Community Edition includes only the basic framework; the advanced TTS and some plugins used by the author during live streams are sold separately in the commercial edition. The Bilibili live stream viewer counts and follower growth figures in the README are self-reported creator dashboard data by the author.
- Live2D Cubism Core is proprietary software; its use and distribution must comply with its license. It has only been tested on Windows.