CompNav · Open-source projects
AIwaifu
A customizable, fine-tunable AI character project for self-hosted companion use, with VTube Studio integration.
What it is
AIwaifu is a customizable AI character project inspired by Neuro-sama. It connects character conversation, speech generation, and a VTube Studio avatar, with a self-hosted and fine-tunable model direction. The repository includes Python code and voice components. It is aimed at users willing to configure the character and services rather than people looking for a hosted chat product.
Core mechanisms and capabilities
Character identity and dialogue can be customized; speech output uses a VITS/Sovits path and can drive VTube Studio. The README describes local language-model responses and a setup that can be extended into a livestream character.
Editorial take
The project has a meaningful community following among open-source AI companion and VTuber projects, and its README directly documents character chat, voice, streaming, and local model setup. It references older model and Python environments, with relatively demanding hardware and manual setup steps; these are important context, not a promise of turnkey use.
Who it suits
Suited to users who want to connect a local model with character dialogue, speech, and VTube Studio and can handle Python dependencies, model downloads, and service configuration.
Deployment and technical requirements
The project separates local inference from the character client. Users install dependencies and models, then configure the API, speech engine, and VTube Studio connection. The README provides setup steps, including manual native voice-module builds and configuration-file changes.
Runtime requirements
The README lists prerequisites including Python 3.8, C/C++ build tools, CMake, Git LFS, and VTube Studio; sample models and speech services may require additional weight downloads. Its hardware references include 12 GB system memory or 8 GB VRAM, with actual performance depending on the model and device.
The project describes self-hosted open models. Users download a model, prepare inference, and configure speech and VTube Studio. The README suggests at least 8 GB VRAM, or CPU inference with slower response times.
Limitations and considerations
The README describes an older default language model and voice pipeline, and some dependencies may need adjustment for current environments. Character assets, model weights, and external software have separate terms; the source code is GPL-2.0. Demonstrated capabilities do not imply that every component runs smoothly on a low-spec computer.
Official sources and further reading
Information checked: Oct 26, 2026.