VIREA

_{Verifiable Interaction Runtime for Embodied Agents · /ˈvɪəriə/}

✦ Bring your avatars to life

VIREA brings your existing avatars to life through language, speech and generated movement. It retargets native skeletal output onto VRM avatars in the browser and Humanoid avatars in VRChat, with a complete path from model setup to generation, playback and inspection.

VIREA-English-Web-v2.mp4

Watch the 4K original · Film editions & editable project

🗓️ October 2026 · source 0.4.0 — Current main · Changelog

01 · 🦴 Motion that fits your rig Source rest frames, joint hierarchies and rotation conventions enter one retargeting foundation. Inspect the original skeleton beside your avatar, including hands, root movement and ground contact. Explore the mathematics →

02 · 🔌 Models you can combine Use MotionCraft, SynTalker, ARDY and other catalog models through adapters for their native output. Connect cloud or local LLMs and TTS through typed capability ports and verify each with a real generation. Meet the models →

03 · 🎬 One complete performance Speech and movement have independent timelines. Prepare an action sequence, place speech where it belongs, and deliver it to the browser or VRChat. Versioned messages, artifacts and execution receipts make each step inspectable. Read the contracts →

---

🧭 Where VIREA fits

Different projects start from companionship, animation authoring or motion capture. VIREA brings generation, retargeting and embodied execution into a shared runtime.

Project	LLM + speech	Generated motion	Native skeleton → VRM	Browser avatar	VRChat execution	Independent tracks	Motion / execution contracts
Project Gabriel	✅	✅	❌	❌	✅	❌	❌
rexclaw	✅	✅¹	❌	✅	❌	❌	❌
text-to-vrma	❌	✅	✅	✅	❌	❌	❌
MotionMind	✅	❌²	✅	✅	❌	❌	❌
ChatVRM	✅	❌	❌	✅	❌	❌	❌
VIREA	✅	✅	✅	✅	✅	✅	✅

_{✅ Implemented in the reviewed scope · ❌ Not provided in that scope. ¹ Through Text-To-VRMA integration. ² MotionMind reconstructs motion from video and composes clips. Independent tracks means explicitly scheduled speech and motion; contracts means both normalized motion and target execution receipts. Reviewed October 11, 2026. Definitions, revisions and sources.}

---

🚀 Set up

Start with a folder and the launcher. Download the source ZIP and extract it, or clone with Git:

git clone https://github.com/Moonweave-AI/virea.git
cd virea

Run the command for your system from the project folder:

Windows · PowerShell	macOS / Linux · Terminal
.\Start-VIREA.cmd	sh start-virea.sh

The launcher prepares Python, Node.js and application dependencies, builds the interface and starts VIREA. The same command starts it again next time.

1. Open Set Up when the launcher is ready.
2. Choose your language, speech and motion providers. Detect available models and voices, or deploy a local model from the catalog.
3. Run generation verification, then enter Motion Lab, Character Studio or VRChat Studio.

Tip

Browser workspaces run on Windows, Linux and macOS. VRChat integration runs on Windows. Local model requirements are shown before installation; start with the capabilities you need.

Full setup guide · Platform guide · Troubleshooting

Workspace	Default address
Set Up · configure and deploy	Open
Motion Lab · generate motion	Open
Character Studio · converse with a character	Open
VRChat Studio · connect the game	Open

For a custom port, use the links printed by the launcher. After updating the code, close the launcher and run it again; settings, models and results stay in the same data folder.

---

🎞️ Demos

Motion Lab

PRISM and SentiAvatar · original skeletons and retargeted avatars, side by side.

virea-motion-lab-current.mp4

Character Studio

A MotionCraft performance with an existing VRM character and an English reply.

virea-character-studio-current.mp4

VRChat Studio

Generated full-body movement and subtitle replies, captured in a VRChat mirror.

virea-vrchat-studio-web.mp4

All demos and reproduction notes · Media provenance

_{Demo avatar: Unnamed Character 6 — Reira. Credits & terms.}

---

🧩 Architecture

One capability system. One motion foundation. Multiple execution targets. The workspaces share model interfaces and skeletal mathematics. Each target handles its own rendering or device transport.

%%{init: {"theme":"base","themeVariables":{"primaryColor":"#302b3a","primaryTextColor":"#f2edf1","primaryBorderColor":"#bdb8d1","secondaryColor":"#352c33","secondaryTextColor":"#f2edf1","secondaryBorderColor":"#c9adb8","tertiaryColor":"#23222b","tertiaryTextColor":"#f2edf1","tertiaryBorderColor":"#a49baa","lineColor":"#a89eb3","textColor":"#716878","edgeLabelBackground":"#eee9ee","clusterBkg":"#23222b","clusterBorder":"#a49baa","titleColor":"#f2edf1","noteBkgColor":"#eee9ee","noteTextColor":"#302b3a","noteBorderColor":"#a49baa"}}}%%
flowchart TB
    subgraph Workspaces["Three workspaces"]
        Lab["Motion Lab"]
        Studio["Character Studio"]
        VRC["VRChat Studio"]
    end
    subgraph Capabilities["Pluggable capabilities"]
        LLM["LLM · intent and planning"]
        Audio["TTS · measured speech"]
        Motion["Motion generators · native outputs"]
    end
    Lab --> Motion
    Studio --> LLM
    VRC --> LLM
    LLM --> Plan["Performance compiler · independent tracks"]
    Plan --> Audio
    Plan --> Motion
    Motion --> Source["Source adapters · Motion IR v2"]
    Source --> Rig["Shared skeletal retargeting<br/>Frames · FK · rotations · hands · contacts"]
    Audio --> Performance["Executable performance"]
    Rig --> Performance
    Performance --> Executor["Session + executor<br/>Timing · ownership · cancellation · receipts"]
    Rig --> Export["Motion Lab · inspect / export"]
    Executor --> Browser["Browser VRM"]
    Executor --> Bridge["OpenVR / OSC → VRChat Humanoid"]
    Browser -. acknowledgements .-> Executor
    Bridge -. acknowledgements .-> Executor
    classDef workspace fill:#23222b,stroke:#b1aabd,color:#f2edf1
    classDef core fill:#302b3a,stroke:#bbb0ca,color:#f2edf1
    classDef target fill:#352c33,stroke:#c7adb7,color:#f5eef0
    class Lab,Studio,VRC workspace
    class Plan,Source,Rig,Performance,Executor core
    class Browser,Bridge,Export target

Loading

Layer	Responsibility	Code
Capabilities	Interchangeable language, audio, motion, source and embodiment interfaces	src/virea/ports/
Contracts & orchestration	Plans, measured timing, immutable artifacts, body ownership and receipts	packages/contracts/ · src/virea/application/
Motion foundation	Native source adapters, rest frames, FK and skeletal retargeting	src/virea/motion/ · packages/retarget/
Execution & operations	Browser playback, VRChat tracking, model workers and deployment	apps/ · src/virea/vrchat/ · plugins/models/

Architecture guide · Timing and execution · Model integration

---

🤝 Contributing

Bring a model adapter, a clearer guide, a reproducible motion example or a fix. Start with the contribution guide, choose a focused change, and include the checks that show it works. For motion changes, an original-skeleton / target-avatar comparison is especially useful.

---

🛡️ Security & responsible use

Use avatars and voice references you have permission to use. Keep provider keys and private conversations local, and operate VRChat accounts with the consent of the people involved. See the security policy for private reporting and the asset credits for model, dataset and avatar terms.

---

📄 License

The VIREA code license is awaiting the maintainers' selection. Third-party models, datasets and avatars retain their own licenses, listed in Third-party notices.