StudioCast

StudioCast is an open-source Linux application that provides a virtual camera, virtual microphone, and virtual speakers with real-time effects for video calls, streaming, and recording.

It's Pretty Awesome NGL

Most Linux broadcast-effect projects prove one cool trick or half bake the broader idea. StudioCast is on another level and constantly improving!

• Broadcast-style virtual camera, virtual mic, and virtual speakers in one app.
• Daemon-first design keeps devices available while heavy processing wakes only when real apps start using them. Also this allows us to run headless... SPOOKY!
• Open backends first, optional Maxine power when you bring your own SDK.
• GPU-conscious pipelines built around efficiency, reuse, fewer transfers, and low-latency real-time calls with CPU fallback!
• Written in C++ with speed in mind!
• Don't like the ML models provided? First off, ouch. Secondly, we made StudioCast bring-your-own-model friendly so feel free to try almost any model you like.
• Safety-minded audio routing that avoids feeding StudioCast back into itself.
• Desktop GUI, CLI tools, diagnostics, installer flow, and service integration instead of a pile of one-off scripts.
• Built for OBS, Zoom, Teams, Discord, and browser calls, with the architecture to grow from hobby setup to daily-driver streaming rig.

What it does

• Creates a virtual camera that can be selected in OBS, Zoom, Teams, Discord, and browser/WebRTC apps.
• Provides virtual audio devices for microphone and speaker routing.
• Runs optional video effects such as background blur/removal/replacement, denoise, relighting, eye contact, auto-framing, and mirroring when the required backend is available.
• Runs optional microphone and speaker effects such as noise removal and voice enhancement when the required backend is available.
• Uses a background daemon, studiocastd, so virtual devices can stay available while heavier processing starts only when an app is actually consuming them.

Project status

StudioCast is an early-preview Linux project. It is usable for testing on Ubuntu 22.04 and 24.04, with a source-build flow and an installer wizard target for release packaging. Packaging, model installation, hardware behavior, and some effects are still evolving.

Known caveats:

• The open-source eye-contact/eye-tracking path is best-effort and can be glitchy.
• Effect availability depends on local GPU drivers, ONNX Runtime, model packs, or an optional NVIDIA Maxine SDK installation.
• Development bandwidth may vary. Focused contributions and good bug reports are welcome.

Requirements

• Ubuntu 22.04 or 24.04, or a close Ubuntu-family desktop distribution.
• A V4L2-compatible physical camera for camera input.
• v4l2loopback for the StudioCast virtual camera.
• A PulseAudio-compatible audio stack with pactl for virtual audio routing.
• CMake, Ninja, Qt, and compiler dependencies for the current source-build flow.
• Optional: NVIDIA driver/CUDA support for Open Video model backends.
• Optional: user-installed NVIDIA Maxine SDK assets for Maxine effects.

StudioCast does not ship NVIDIA Maxine SDK files or model binaries in git.

Install options

• Download GUI installer: release packaging builds a versioned StudioCast-Installer-<version>-<arch>.AppImage when AppImage tooling is available, plus a staged AppDir archive, standalone source archive, and SHA256 checksums. Run the installer as your normal user, not with sudo. The AppImage/AppDir is self-contained for the installer GUI, scriptable backend, and matching StudioCast source archive; the GUI preselects that bundled archive for install/update workflows. Runtime dependencies still come from supported system packages and the backend setup scripts. The installer builds StudioCast on the target system; it is not a binary-only StudioCast application bundle.

• Build the GUI installer package locally:

packaging/appimage/build_appimage.sh

If linuxdeploy and linuxdeploy-plugin-qt are installed, this also produces an AppImage. Without those tools, it produces a staged AppDir tarball and checksum for inspection/testing.

• Build the GUI installer target from source:

cmake -S . -B build -G Ninja -DCMAKE_BUILD_TYPE=Release
cmake --build build --target studiocast-installer
./build/studiocast-installer

• Build from source manually: use the commands below when developing, installing over SSH, debugging setup, or recovering from a failed install.

The same backend used by the GUI is available as a CLI fallback:

./scripts/installer.sh status
./scripts/installer.sh plan install
./scripts/installer.sh install --yes
./scripts/installer.sh uninstall --yes

Quick start from source

Run these commands from the repository root.

1. Install system dependencies and create the virtual camera:

./scripts/setup.sh --deps --v4l2loopback --load-loopback --persist-loopback

By default this creates a v4l2loopback device such as /dev/video10 labeled StudioCast Camera.

2. Build StudioCast:

./scripts/setup.sh --build --build-dir ./build --build-type Release

3. Start the daemon:

./build/studiocastd

4. In another terminal, check status and open the GUI:

./build/studiocastctl status --pretty
./build/studiocast

Optional: once the manual run works, install the systemd user service:

./scripts/install.sh user-service --build-dir ./build --yes
systemctl --user status studiocastd.service

More setup detail is available in docs/SETUP.md.

Using StudioCast

Start studiocastd, open the StudioCast GUI, then choose StudioCast devices in the app you want to use.

• OBS: add a Video Capture Device and select StudioCast Camera.
• Zoom, Teams, Discord, and browser/WebRTC apps: select StudioCast Camera in camera settings. If the app already had its device list open, refresh or reopen the settings page.
• Audio apps: when StudioCast virtual audio devices are enabled, select StudioCast Microphone and/or StudioCast Speakers in the target app.
• GUI preview: enabling preview can act as a consumer of the virtual camera. Leave it off when you only want external apps to consume the output.

StudioCast is designed to idle when no consumer is using the virtual camera or processed audio path. If it stays busy while no app is consuming StudioCast devices, please file a bug with a support bundle.

Models and effects

StudioCast can use multiple effect backends. The GUI and studiocastctl status report which engines are available on the current machine.

• Open Video / Open CUDA: ONNX Runtime with CUDA execution provider and user-installed model packs. See docs/open_source_video_models_install.md.
• Open Audio: ONNX Runtime and user-installed model packs for microphone effects. See docs/open_source_audio_models_install.md.
• NVIDIA Maxine: optional user-installed NVIDIA SDKs and feature packs. See docs/maxine_install.md.

Curated model installers are available through:

./scripts/install.sh open-video-models --list
./scripts/install.sh open-audio-models --list

Use the model helper to inspect installed packs:

./build/studiocast-open video-list-models
./build/studiocast-open audio-list-models

Troubleshooting

Virtual camera does not appear:

v4l2-ctl --list-devices
./scripts/setup.sh --v4l2loopback --load-loopback --persist-loopback

Daemon is not reachable:

./build/studiocastd
./build/studiocastctl status --pretty

If using the user service:

systemctl --user status studiocastd.service
journalctl --user -u studiocastd.service -f

Apps show the StudioCast camera but no video:

• Confirm the daemon is running.
• Confirm a readable physical camera is selected or available for auto-select.
• Check ./build/studiocastctl status --pretty for device, consumer, and pipeline state.
• Try OBS first when debugging, then browser/WebRTC apps.

Effects are unavailable:

• Check ./build/studiocastctl status --pretty for engine diagnostics.
• For Open Video, verify NVIDIA driver/CUDA, ONNX Runtime, and model packs.
• For Open Audio, verify ONNX Runtime and model packs.
• For Maxine, run ./build/studiocast-maxine install-hints.

Audio devices do not work:

• Check pactl info, pactl list short sources, and pactl list short sinks.
• In the GUI, select a physical microphone or speaker device rather than a StudioCast virtual device as the processing source/target.

Support bundle:

./build/studiocastctl debug-report --out studiocast-debug-report.txt

Attach the report when opening an issue.

Contributing

Start with docs/DEVELOPER_GUIDE.md for build, architecture, daemon, IPC, model, and testing notes.

Project conventions are in CONTRIBUTING.md. Please also see CODE_OF_CONDUCT.md and SECURITY.md.

License and trademarks

StudioCast is independent and is not affiliated with, endorsed by, or sponsored by NVIDIA. NVIDIA and NVIDIA Broadcast are trademarks of NVIDIA Corporation. StudioCast does not ship or redistribute NVIDIA Broadcast or Maxine SDK binaries.

See LICENSE, NOTICE, and docs/TRADEMARKS.md.