VTube Studio is the most common bridge between a Live2D avatar and your stream, and you can get a model moving on screen in about 20 minutes. The workflow is simple once you know the order: load the model, connect tracking, tune the sensitivity, then pipe the transparent output into OBS. This walkthrough covers a PC-and-webcam setup and the optional iPhone tracking path, plus the settings that make an avatar look intentional rather than twitchy.
What you need before you start
You need a Live2D model file (a .vtube or model3.json folder), VTube Studio installed on your PC, a webcam or a Face ID iPhone for tracking, and OBS for the actual broadcast. Free starter models ship with the app, so you can practice the whole pipeline before commissioning a custom avatar that typically runs $150–800 depending on rigging complexity.
Step 1: Load your model
Open VTube Studio, click the model icon, and select your avatar. It appears centered on a colored background. Drag to reposition, scroll to scale, and set an idle animation if the rig includes one. Spend a minute here so the framing matches where you want the avatar to sit in your overlay — usually bottom-right or bottom-left over gameplay.
Step 2: Connect face tracking
For webcam tracking, open the settings gear and pick your camera under the tracking source. For iPhone tracking, install the VTube Studio companion app on the phone, put both devices on the same Wi-Fi, and enter the IP address shown on the phone. iPhone ARKit gives you far richer mouth and blink data; a webcam is perfectly usable but simpler.
Calibrate a neutral face
Sit in your normal streaming posture, look at the camera, and trigger the calibration so a relaxed expression maps to the avatar’s neutral pose. Recalibrate any time you change your chair height or lighting.
Step 3: Tune the tracking parameters
| Parameter | Starting value | Raise it if… | Lower it if… |
|---|---|---|---|
| Smoothing | 0.4–0.5 | Avatar jitters | Avatar feels laggy |
| Mouth sensitivity | Default +10% | Mouth barely opens | Mouth flaps constantly |
| Eye blink threshold | Default | Blinks get missed | Avatar squints often |
| Head angle range | 0.8× | Turns feel stiff | Head swings too far |
Adjust one slider at a time and watch your face in the preview. The goal is that a small real movement produces a small, believable avatar movement, with a slight boost so subtle expressions still read on camera.
Step 4: Send a transparent feed to OBS
In VTube Studio settings, enable Spout2 output (Windows) or use a colored chroma background you key out. Spout2 is cleaner because it carries true transparency with no green-screen edges. In OBS, add a Spout2 Capture source, select VTube Studio, and your avatar drops in with a transparent background ready to layer over any scene.
Add hotkeys and expressions
Once the model tracks well, map hotkeys to expressions and props so you can trigger a blush, an angry pop, or a prop appearing with a keypress or a Stream Deck button. In VTube Studio, open the hotkey settings, assign an expression or animation, and bind it to a key. Keep it to a handful you will actually remember mid-stream — three to six is plenty to start. These triggered expressions are what make a VTuber feel reactive and alive, turning a talking model into a performer. Test each one on a private stream so the timing and placement feel natural before you use them live.
FAQ
Why is my avatar’s mouth not moving with my voice?
VTube Studio tracks mouth shape from the camera, not audio, by default. Make sure your mouth is visible and well lit, raise mouth sensitivity, and recalibrate. If you want audio-driven lip sync, enable the microphone-based mouth tracking option and select your mic.
Can I run VTube Studio and a game on the same PC?
Yes, but tracking and rendering add roughly 5–15% CPU/GPU load. On a mid-range system that is fine for most games; on a tight build, lower the tracking frame rate to 30fps or offload tracking to an iPhone to free the PC.
Bottom line
Load the model, connect tracking, tune smoothing and mouth sensitivity a little above default, then send a Spout2 transparent feed into OBS. Start with a free model to learn the pipeline, and only commission a custom rig once you know exactly how you want it to move. Twenty minutes of calibration is the difference between an avatar that looks alive and one that looks like it is buffering.
