VTuber models: PNGTuber, Live2D, 3D and OBS setup

PNG, Live2D or 3D: start with the performance

A static PNG can represent you in a scene without facial tracking. Live2D animates prepared parts of an illustration. A 3D avatar adds a spatial representation; VRM standardizes elements such as a humanoid skeleton, expressions and gaze for compatible applications.

For a talking stream, prioritize the expressions your character needs. For dance or broad movement, also plan how the body will be captured. A file format alone supplies neither sensors nor a performance: ask to see the use case you actually want.

How to start as a PNGTuber with veadotube mini and OBS

For a talking avatar without facial tracking, try a PNGTuber before commissioning a rig. This test route uses veadotube mini, a separate application from veadotube. Use artwork you own or have permission to broadcast.

1. Add closed-mouth and open-mouth images to a selected state on the right. The file menu loads .veado avatars, not individual images.
2. Select your microphone. Adjust the green activation threshold while speaking, then leave a quiet pause; use the purple control to adjust the return delay.
3. Capture the avatar in OBS using the method documented for your operating system; check the background and crop.
4. Save a .veado backup. Record a short local test and listen back: a moving mouth does not establish that OBS recorded your microphone.

PNGTuber mouth, transparent background or voice missing: what to check

On Windows, Game Capture supports transparency: enable Allow Transparency when mini uses a transparent background. On macOS and Linux, follow their separate routes in the official guide. Window Capture may need a contrasting background and a Color Key filter. Test the full scene on your own hardware.

Design, illustration and rigging are different jobs

Character design establishes the character; illustration prepares its appearance. For Live2D, separating eyes, mouth, hair and other parts makes movement possible, including artwork normally hidden behind another part. Rigging defines how those parts change: Cubism parameters can represent mouth opening or a head angle.

One person may do several jobs, but agree on the responsibilities separately. Ask who prepares layers, who rigs the model and who will investigate a defect after import. Community labels such as mama and papa do not replace role-specific credits.

Which files need to be delivered?

In Cubism, .cmo3 is the model project; .moc3 contains runtime model data. A .model3.json file references the model, textures and other settings. A runtime file is not the editable project.

VTube Studio imports the exported package and its associated files. Its documentation says to export physics together with the model so that it is registered. Before accepting delivery, test the complete folder in the agreed application; keep an original copy and record the version used.

For 3D, distinguish VRoid Studio’s model editor from its Dress-up feature. The editor saves .vroid and offers VRM 0.0 or 1.0 export; Dress-up uses XRoid data to retain outfit adjustments and exports VRM 1.0. Accepting a VRM in Dress-up does not mean recovering its original .vroid project. The .vrm extension alone does not tell you which version your destination application supports.

Back up a VTuber model and OBS, then test the restore

Exporting an OBS scene collection saves JSON data; it does not package the images or videos referenced by its sources. Keep those assets separately. An editable project, the model opened by your tracking app and your OBS configuration are also different deliverables.

Before changing computers or reinstalling, prepare a dated copy outside the working computer and a list of required applications, versions and assets. Retain the editable originals included in your commission. Treat the backup as private: keep account settings, keys and private widget links out of public support bundles.

1. Inventory first: model, agreed editable files, illustrations, videos, sounds, plugins and their versions. Record where each asset is used. Do not assume an export installs plugins.
2. Close applications before copying their working folders. Keep the last working backup intact; restore to a separate location and follow your version’s import instructions. Do not overwrite the only copy while testing.
3. Test in layers: open the model, exercise tracking and expressions, then check OBS scenes. If an image or sound is missing, check its file and path before rebuilding the rig.
4. Make a local recording and listen to voice and game sound. Check the transitions and hotkeys you use. A visible preview does not prove that the complete recording setup or streaming connection has been restored.

VTube Studio or VSeeFace: which app opens your model?

Choose the application before commissioning or exporting the model. Check the file it opens separately from the tracking data it receives: a phone sending facial movements does not change which avatar formats the receiving application supports.

VTube Studio: connect iPhone or Android face tracking to PC over Wi-Fi

The phone sends facial movements to desktop VTube Studio, which displays the Live2D model. You do not need the model open on the phone too. Capturing the avatar and recording voice in OBS are separate steps.

If Find Server cannot see the computer, check the shared local network, router or VPN device isolation, and the VTube Studio firewall exception. Keeping the computer’s protection permanently disabled is unnecessary.

1. Open VTube Studio on phone and computer on the same local network. The computer can use Ethernet to the same router.
2. Enable the server in desktop settings. Try the default port 25565 first and allow VTube Studio on the trusted private network.
3. On the phone, use Find Server or enter the IP and port shown on desktop, then choose Connect to PC.
4. Load the desktop model and check its response to your face before moving to OBS. If Updates stays at 0, the official guide suggests restarting the app.

iPhone USB tracking: connection order and device not detected

The current guide includes Apple Devices as an alternative to iTunes for USB communication on Windows. If you choose iTunes, it says to open it at least once to start the communication service. This prepares the connection; it does not add the iOS direct-USB mode to Android.

1. Connect the iPhone or iPad to the computer by USB, unlock it, open VTube Studio, and complete the computer-trust prompt if shown.
2. Enable Connect USB on the phone first, then in desktop VTube Studio. This mode replaces the Wi-Fi tracking connection.
3. If the device is absent, check the cable, unlocked state, trust pairing and Apple communication service. Look for USB started. Waiting for PC app to connect. on the phone. Changing the model format does not resolve USB device detection.

If the model will not load: check format before tracking

Separate three results: the avatar appears, it responds to tracking, and OBS captures the image. If the first fails, changing the camera or OBS scene does not establish file compatibility. Keep the original and record the application, version, exporter and exact error.

1. Identify the delivery. .cmo3 and .vroid are editable projects: return to the appropriate editor to export the agreed runtime package, rather than changing the extension.
2. If VSeeFace reports NotVrm0Exception, its FAQ says VRM 1.0 is unsupported. Check the actual .vrm export format before investigating tracking.
3. Return to VRoid’s model editor and choose a compatible export. Dress-up exports VRM 1.0, not 0.x: if that is your only delivery, choose a compatible application or arrange another delivery with the creator; renaming the file does not convert it.
4. Try a compatible sample model in the same application, followed by your delivery. If only your file fails, record the difference for the creator to investigate; if both fail, check the application installation and logs before rebuilding the model.

VTube Studio model not moving? Check tracking input and model output

Rigging prepares movements the model can display. Tracking generates signals from a camera or another device; the application applies them to the model. More input data does not automatically create an expression that the model has not been built to show.

VTube Studio supports Live2D models with webcam or compatible mobile face tracking. It does not load VRM/3D avatars itself, although it documents integrations that send tracking data to other applications. Confirm the device, target application and model format before buying or commissioning.

With the model loaded, open VTS Parameter Setup in its settings. Observe the tracking INPUT value, the Live2D parameter OUTPUT value and the visible character movement separately. Automatic blinking or breathing can run without tracking: seeing them does not prove that camera data is arriving.

1. Save a copy of the model configuration and record current settings before changing them; do not delete the rig or delivered files to test a hypothesis.
2. Test one movement at a time and compare a compatible sample model using the same device. A difference helps locate the problem but does not by itself prove a fault in the commissioned model.
3. If you need the creator’s help, provide the VTube Studio version, device, observed INPUT and OUTPUT values, their ranges and active expressions. Describe the expected movement and the result of each test.

VRoid commercial use, VRM permissions and commission rights

Can you monetize a VRoid avatar? Yes: the created-model FAQ permits commercial use subject to the applicable terms. The software FAQ covers a different scope: use of the editor. Do not read its restrictions as a blanket ban on monetizing models.

The VRM 1.0 specification includes metadata for avatar permissions, commercial use, attribution, redistribution and modifications, with possible additional terms. Check the particular avatar’s values; the VRM format does not itself grant every permission.

For a commission, request written terms for the planned uses and included files: monetized streams, videos, merchandise, future edits and credits. Receiving an editable project and having permission to modify it are different questions. Do not infer the terms from a public showcase.

1. List the model, textures and accessories; keep their creators and applicable terms together.
2. Record your intended uses and required credits. Resolve unclear permissions with the rights holder before publishing or commissioning edits.
3. Check the delivered file’s permissions and additional terms. Changing metadata yourself does not grant new rights.

VTube Studio expression setup: from model to hotkey

Test the expression directly before troubleshooting its shortcut. Missing rigging needs the model creator’s input.

1. Load the model. In Hotkeys, open Expression Editor and manually switch one expression on, then off.
2. Create a hotkey with Set/Unset Expression and select that expression’s file.
3. Assign keys or an on-screen button; check both activation and deactivation.
4. Record a short OBS test with the game in focus. Check the saved recording.

VTube Studio hotkeys not working: four checks

Change one thing at a time. An active expression can override tracking; Add and Multiply also alter the result. Disable expressions for comparison before recalibrating.

Expressions marked Model Customization Expression survive Clear all Expressions. Switch those off individually for a full comparison.

Your first VTuber stream in OBS: setup and test

OBS combines sources into scenes, including images, application captures, video and audio. The avatar application produces the character image; the scene combines it with the rest of the broadcast. Adding a capture does not make OBS a rigging editor.

Start with a short recording: check cropping, background, voice volume and whether private windows appear. Then test with the game or application you actually plan to use. This checks the combined result rather than the avatar in isolation.

1. Save your current profile and scene collection before changing settings. In Tools → Auto-Configuration Wizard, choose the streaming route. Review the result in Settings → Stream, Output and Video: it is a starting point for your computer and connection, not a universal preset.
2. Select the destination service and account. If you use a stream key, keep it outside the capture. Before clicking Start Streaming, confirm the platform’s test mode or visibility; a local recording does not verify the upload to its server.
3. Test the output with the avatar, voice and game you plan to use. Watch OBS Stats and the platform dashboard. Afterwards, stop sending and confirm there that the test has ended. Save the profile, settings and result so you can compare the next change.

VTuber and OBS lag: rendering, encoding or network?

Separate choppy tracking from slow animation or a recording that loses smoothness. VTube Studio’s troubleshooting guide discusses GPU competition with games and performance options. Replacing a camera does not necessarily solve an overloaded GPU.

Compare recordings with and without the game, change one setting at a time and note the result. There is no useful blanket guarantee that every 2D model is lighter than every 3D model: evaluate the actual model, applications and scene. A concert-studio setup is not a prerequisite for a talking stream.

1. Save the current profile and scene collection. Note FPS, output resolution, encoder and bitrate; compare the same short sequence with the avatar alone, then with the game.
2. Compare counter increases during each test, rather than their lifetime totals. Change one option, repeat and keep the comparison. Several categories can worsen together; there is no universal preset.
3. Compare the local file with stream playback. Network counters require an actual transmission; zero losses do not certify the source or every viewer. If problems persist, keep that session’s log for OBS support.

Delivery test: what to do and how to locate a fault

This USADA procedure combines documented application features with a practical check. It is not an official certification or a performance guarantee: agree on the expressions, equipment and expected result with your model maker.

1. Import a complete copy of the delivery into the agreed application and record the file and app versions. It should open with its textures and associated settings. If it does not load, check the package and compatibility first; camera calibration cannot repair missing files.
2. In VTube Studio, switch each agreed expression on and off from the Hotkeys tab, then test its shortcut. Check the change and the return to the normal state. If the tab control fails, investigate the delivered expression; if only the shortcut fails, check its binding before attributing the problem to rigging.
3. Next test tracking. For a webcam in VTube Studio, calibrate while looking straight ahead with a neutral face; test blinking, mouth opening and head turns separately. If manual controls work but your face does not drive the expected movement, check input, calibration and parameter mapping before concluding that the movement is missing.
4. Make a local OBS recording with the intended scene and compare it with the avatar application window. Expressions and framing should carry through. If they look right in the app but wrong in the recording, investigate the OBS source and scene before requesting a model change.
5. Repeat the same sequence with your usual game or other applications running and compare it with the unloaded test. If it only deteriorates under load, investigate resources and settings. Keep both recordings and identify the failing step: they help describe the fault without confusing delivery, tracking and broadcasting.

How to capture VTube Studio in OBS with transparency

Keep the avatar and background as separate sources, so changing the scene does not require changing the model. These routes concern VTube Studio, not every 3D application. Start with a local recording of a duplicate scene.

For Spout2, VTube Studio recommends Premultiplied Alpha with transparent black. Check translucent hair and effects over both light and dark backgrounds; do not resize the whole canvas to fix one source.

Black background, missing colors or lag: what to test first

Compare three places: the avatar window, OBS preview and saved recording. This USADA sequence helps locate the first failing stage; it is not hardware certification or a frame-rate guarantee.

1. If the face already moves incorrectly in the avatar window, return to tracking and calibration. Do not start by changing the OBS encoder: you have not yet established that the input animation is correct.
2. If the avatar works but is absent from OBS, check the selected window or sender, source visibility and scene position. On multi-GPU Windows systems, also check that VTube Studio and OBS use the same dedicated GPU.
3. For a black rectangle or missing colors, check transparency, compositing and filters before rigging. Compare an unkeyed duplicate with the filtered scene to identify whether the filter removes character details.
4. If opening a game causes the problem, compare a simple scene with the usual one. OBS also needs GPU resources: cap the game frame rate and try fewer sources or effects before attributing every slowdown to the model.
5. If the local recording is smooth but the broadcast is not, investigate streaming output separately. If both fail, try a lower output resolution or frame rate and save each result. Change one setting at a time.

No microphone in OBS: test voice separately from the avatar

A moving avatar mouth does not establish that OBS receives your voice. Hearing yourself in headphones is not enough either: check the input, its mix and the recorded file. This USADA sequence helps identify where sound is lost.

1. On Windows or macOS, select the microphone in Settings → Audio or add Audio Input Capture to the scene. If you add it as a scene source, disable global capture of that same device; do not capture it twice.
2. With the game quiet, speak and watch the microphone meter in Audio Mixer. If it does not respond, check the selected device and its input level before changing the model or rigging.
3. Next test the game separately. Balance voice and game by listening to both, not just comparing meter heights. Check that the intended sources are not muted.
4. Record three segments: voice alone, game alone and both together. Play the file and listen to both headphone channels. If a source is missing, check the tracks before concluding that it was not recorded.

Capture game audio without the whole desktop: Windows and macOS

Image capture and audio capture are separate settings. Choose the route supported by your operating system and OBS version; Windows options are not Mac instructions.

On the Windows per-application route, disable global Desktop Audio to avoid capturing the same sound again. Test the chosen application before broadcasting: isolating an app does not exclude sounds or alerts generated inside it.

Separate tracks: why a recording plays the game but not your voice

A player normally plays one audio track at a time rather than mixing every track in the file. To edit voice and game separately while retaining an easy listening mix, you can use the example below; it is not an automatic OBS assignment.

In Audio Mixer → Advanced Audio Properties, assign sources to their tracks. Then enable the tracks to record in Settings → Output. Check both the mix and the separate tracks in your player or editor; assigning a source to a track is not enough if that track is not recorded.

Twitch VOD audio: keep voice and game in the archive

OBS can send different mixes to the Twitch live stream and VOD. The official guide defaults to track 1 for live audio and track 2 for the VOD. Configure this in Output after connecting Twitch; controls depend on output mode.

This example replaces the previous one, where track 2 contains only voice. Excluding a source from the VOD does not authorize its live use.

1. Save the current configuration. Use your voice and self-made sounds as recognizable markers.
2. Assign sources in Advanced Audio Properties; avoid capturing them again through Desktop Audio.
3. Enable the chosen local tracks and play them separately. A local file does not prove the VOD mix.
4. With VOD storage enabled, plan a short broadcast that may be public. Afterwards, check voice and game in the Twitch archive; a bandwidth test does not create that VOD.

Missing voice, echo or one-sided audio: locate the symptom

Change one route at a time and keep the test recording. These checks distinguish input, capture, mixing and playback; they do not require buying another microphone or rebuilding the avatar first.

Budget: compare the same commission, not just the deposit

Two “complete model” quotes can cover different work. Request both against the same list of movements, expressions, files and permissions. If one covers artwork alone and another includes rigging, ask for the missing cost; an unpriced line is not a free service.

For planning, separate launch costs, recurring payments over your chosen number of months and a reserve for changes. Add only the options you will buy. A deposit normally forms part of the agreed total: request the payment schedule and do not count it twice. Compare in one currency and clarify taxes, payment fees and quote expiry.

What you can try for free, and what to budget separately

Test your chosen workflow on your equipment first. A free application does not include commissioned artwork, sensors or permission to use someone else’s assets. You also do not need to buy a rigging editor just to load the exported model package supplied by your maker into VTube Studio.

A practical commission brief

Instead of requesting only “a VTuber model,” prepare these decisions and ask for a demonstration against the agreed requirements. This is a planning checklist, not a fixed-price package.

• Main use: talking, singing, gaming or body movement.
• Model format, target application and tracking device you already have.
• Required movements and expressions, with clear references.
• Responsibility for design, artwork, rigging and import.
• Delivered runtime files, editable projects and application versions.
• Permissions, credits, included revisions and support after delivery.
• A recorded acceptance test with the intended model, scene and workload.