NVIDIA Broadcast
for Linux & macOS

AI-powered virtual camera with background removal, eye contact correction, face relighting, recording, AI meeting transcription, voice studio, and video enhancement. 9 GPU/CPU modes. Linux and macOS.

Install Now View on GitHub Sponsor
Terminal
$
Features

Look professional. Sound professional.

Works with Zoom, Chrome, OBS, Discord, FaceTime. One app does it all.

Blur your background

3 AI models. Real-time.

Sound like a studio

Bass, warmth, compression. GPU.

AI writes your notes

Whisper + summarizer. Local.

Background Removal

Blur, replace, or remove. Three AI models to choose from.

Eye Contact

Look at the camera even when reading your screen.

Face Relighting

Auto-match your face lighting to the background.

Video Enhancement

Skin smooth, denoise, sharpen. Four beauty presets.

Auto-Frame

AI keeps you centered when you move around.

Virtual Camera

Shows up in Zoom, Chrome, OBS, Discord automatically.

Voice Studio

Bass, treble, warmth, compression. 6 presets. GPU-accelerated.

Noise Removal

AI removes background noise from your mic in real-time.

Mic Test + VU Meter

Record 5 seconds, play it back. See your live audio level.

AI Meeting Notes

Whisper transcribes. AI extracts action items. 100% local.

Recording + NVENC

Record video + audio to MP4. Hardware encoding. One click.

Preview

GTK4 + Adwaita, built for Linux

Dark-themed native interface with resizable preview, real-time controls, and status monitoring.

NV Broadcast
REC
ZEUS MODE
33 fps
Processing
Zeus (480p + CuPy)
Camera
EMEET SmartCam S600
Resolution
1920 x 1080 @ 30fps
Background
Blur (Strength: 70%)
Broadcasting to /dev/video10
GPU: RTX 3080 | VRAM: 1.2 GB
How It Works

Camera to virtual output in milliseconds

GStreamer captures, ONNX Runtime infers, CuPy composites, v4l2loopback outputs.

Camera
MJPEG capture
→
GStreamer
Decode + appsink
→
AI Inference
ONNX + CUDA
→
Composite
Fused CUDA kernel
→
Virtual Cam
v4l2loopback
Performance

9 modes, from killer to potato

Real benchmark numbers at 1080p. 5 built-in profiles (Meeting, Streaming, Presentation, Gaming, Clean) plus custom profiles.

Killer
360p + Fused CUDA
48
fps @ 1080p
20 ms frame time
TensorRT + Fused kernel
GPU
Zeus
480p + CuPy
33
fps @ 1080p
30 ms frame time
TensorRT optimized
GPU
DocZeus
720p + Fused CUDA
23
fps @ 1080p
44 ms frame time
Max quality fused
GPU
CUDA Max
720p Quality
22
fps @ 1080p
45 ms frame time
Full resolution inference
GPU
CUDA Balanced
720p skip 2
34
fps @ 1080p
29 ms avg
67% effects ratio
GPU
CUDA Perf
720p skip 2
33
fps @ 1080p
30 ms avg
50% effects ratio
GPU
CPU Quality
720p OpenCV
15
fps @ 1080p
66 ms frame time
SIMD optimized
CPU
CPU Light
720p skip 2
33
fps @ 1080p
30 ms avg
Frame reuse
CPU
CPU Low End
720p skip 3
37
fps @ 1080p
27 ms avg
Maximum skip
CPU
Install

One command, any distro

Detecting your system...
bash
$ git clone https://github.com/Hkshoonya/nvidia-broadcast-linux.git
$ cd nvidia-broadcast-linux
$ ./install.sh

Works on all distros. The installer auto-detects your package manager and installs dependencies. Re-running `./install.sh` upgrades the existing install in place.

bash
# Download the .deb from GitHub Releases
$ wget https://github.com/Hkshoonya/nvidia-broadcast-linux/releases/latest/download/nvbroadcast_1.3.0-1_all.deb
$ sudo dpkg -i nvbroadcast_1.3.0-1_all.deb
$ sudo apt-get install -f
# Upgrade later: repeat the same three commands with the newer .deb

For Ubuntu, Debian, Pop!_OS, Linux Mint, and derivatives. Running the same command with a newer `.deb` upgrades in place. View all releases

bash
# Download the .rpm from GitHub Releases
$ wget https://github.com/Hkshoonya/nvidia-broadcast-linux/releases/latest/download/nvbroadcast-1.3.0-1.noarch.rpm
$ sudo dnf install ./nvbroadcast-1.3.0-1.noarch.rpm
# Upgrade later: download the newer .rpm and run the same dnf install command

For Fedora, RHEL, CentOS, Rocky, openSUSE. Installing a newer `.rpm` upgrades in place. View all releases

bash
# Snap Store (published from Git tags)
$ sudo snap install nvbroadcast
$ sudo snap refresh nvbroadcast

For systems with Snap support. Tag pushes publish the current stable build to the Snap Store, and `snap refresh nvbroadcast` upgrades in place. Users on older builds should refresh to `1.3.0` for the device-resident GPU video path, stronger blur controls, safe camera and microphone power saving, corrected Apple Silicon camera selection, and cleaner sunlit replacement edges. Linux arm64 omits MediaPipe-dependent face effects until upstream provides a wheel compatible with patched Protobuf; background effects, virtual devices, recording, and meeting tools remain available.

bash
# Install system deps then run the installer
$ sudo pacman -S python gtk4 libadwaita gstreamer gst-plugins-base gst-plugins-good gst-plugins-bad v4l-utils
$ git clone https://github.com/Hkshoonya/nvidia-broadcast-linux.git && cd nvidia-broadcast-linux && ./install.sh

For Arch, Manjaro, EndeavourOS. Arch wiki: v4l2loopback

zsh
# macOS (Apple Silicon)
$ git clone https://github.com/Hkshoonya/nvidia-broadcast-linux.git
$ cd nvidia-broadcast-linux
$ ./install_macos.sh
# Upgrade later: install the newer .pkg after this one-time source setup

Existing configured installs can download the update package: NVBroadcast-1.3.0-1.pkg
Requires Apple Silicon, macOS 13+, Homebrew, Python 3.11-3.13, and OBS Studio. After installing OBS, open it once, start and stop Virtual Camera, then close OBS. Intel macOS is not included in v1.3.0 because a secure current MediaPipe wheel is unavailable.

Linux

  • Any distro (Ubuntu, Fedora, Arch, Pop!_OS...)
  • NVIDIA GTX 1060+ for GPU modes
  • Python 3.11+, GStreamer, GTK4
  • PipeWire for virtual mic

macOS

  • macOS 13 Ventura or newer
  • Apple Silicon (M1+) required
  • Homebrew, Python 3.11-3.13
  • OBS Studio for virtual-camera output
  • CPU modes (no NVIDIA needed)

Camera + Audio

  • USB webcam, 720p+ recommended
  • Any microphone (USB, built-in)
  • 360p to 4K supported