OpenVT streaming setup#

This setup is for a simple Linux-first OpenVT pipeline that is ready to render a model window for capture/streaming.

Reality check#

OpenVT is not a trivial single-binary install:

it is built in Godot
it uses third-party dependencies and submodules
full Live2D functionality depends on proprietary Live2D Cubism components
for face tracking, OpenSeeFace is a separate executable

So the first goal here is not full unattended bot control. The first goal is a reproducible local setup path that gets OpenVT rendering and capturable for streaming.

Recommended first streaming path#

Run OpenVT locally on Linux
Load a compatible model
Use transparent window capture in OBS
Stream from OBS to stream.place via WHIP/RTMP
Add bot/chat/TTS control afterward

Requirements#

Godot 4.6+ according to the project docs
OpenVT source tree
built third-party dependencies
optional: OpenSeeFace for webcam tracking
optional: OBS for local production

Local build notes for this repo#

This repo includes a helper for building as far as possible in a Nix/dev shell environment:

devenv shell -- ./scripts/openvt-build-local.sh

It works around some upstream assumptions:

upstream expects pyenv
upstream tries to pip install tools globally
upstream keylogger build wants Rust targets and libudev

This still does not bypass the Cubism SDK requirement.

Notes from upstream docs#

OpenVT focuses on Linux availability
supported trackers: OpenSeeFace and VTubeStudio mobile TCP over Wi-Fi
transparent window support is one of its key streaming features
hotkey capture on Linux may require the input group

Hotkey note#

If OpenVT hotkeys do not work under Linux while unfocused, upstream says you may need:

sudo -g input -u "$USER" open_vt.x86_64

Or to permanently add the user to the input group.

Next integration step#

Once OpenVT is rendering reliably, we can layer on:

chat-triggered expressions
TTS-driven mouth movement heuristics
stream.place chat read/write
autonomous reply logic

Configure Feed