FACELESS
HOW IT WORKS FEATURES TECH PERFORMANCE DOWNLOAD
VERSION 1.0 — GPU ACCELERATED

Your face. Their face. No one's face. 30fps.

GPU-accelerated real-time face swapping at 30fps+. CUDA-optimized pipeline. Electron UI. Virtual camera output. Become anyone.

DOWNLOAD FOR WINDOWS VIEW SOURCE
SOURCE FACE
FACE_REF_001.JPG
Drop image or click to select
31.2 FPS
LIVE CUDA:0 RTX 4070 • 854x480 → 1920x1080
CONTROLS
QUALITY PRESET
NORMAL
HIGH
FACE SWAP
Enabled
GPEN ENHANCE
BFR-256
VIRTUAL CAMERA
OBS VCam
MANY FACES
Off
COLOR CORRECT
On
OPACITY
85%
SHARPNESS
0.4
01 // THE MASK

Become Anyone in Real Time

Four stages in the pipeline. From source face to virtual camera. The mask applies itself.

01
SELECT
Choose a source face from any image. The neural network extracts and maps the face embedding.
02
CAPTURE
Camera feed captured at 1080p/30fps. RetinaFace detects faces asynchronously in real time.
03
SWAP
inswapper_128 replaces faces at 480p. GPEN-BFR-256 enhances. Upscaled back to 1080p.
04
OUTPUT
Feed to Discord, Zoom, OBS via virtual camera. Live preview in the Electron UI.
02 // FEATURES

The Engine Under the Mask

Every tool engineered for speed. Every feature built for the real-time pipeline.

CUDA GPU ACCELERATION
ONNX Runtime CUDA Execution Provider. Full GPU-accelerated inference pipeline on NVIDIA RTX hardware.
RESOLUTION-DECOUPLED
Capture at 1080p, process at 480p, output at 1080p. Eliminates the biggest bottleneck while keeping quality.
GPEN FACE ENHANCEMENT
GPEN-BFR-256 neural face restoration. Adds ~5ms per frame for dramatically improved face quality.
VIRTUAL CAMERA OUTPUT
Feed swapped video to Discord, Zoom, OBS, or any app via pyvirtualcam. Seamless integration.
MULTI-FACE SUPPORT
Swap multiple faces simultaneously in the same frame. Every detected face gets processed.
QUALITY PRESETS
NORMAL mode for balanced 30fps. HIGH mode for maximum quality. Tune for your hardware.
POST-PROCESSING SUITE
Color correction, mouth masking, sharpening, opacity control. Fine-tune every aspect of the swap.
WEBSOCKET ARCHITECTURE
Decoupled Python backend + Electron frontend. JSON control messages. Binary JPEG frame streaming.
03 // TECH STACK

Forged in CUDA Fire

A decoupled architecture. Python inference backend. React Electron frontend. Connected by WebSockets.

Python 3.12
ONNX Runtime CUDA
React 19
Electron
TypeScript
Tailwind CSS 4
InsightFace
WebSockets
Zustand
ARCHITECTURE
ELECTRON UI (React 19 + TypeScript)
↓
WebSocket (ws://127.0.0.1:7865)
↓
Python WebSocket Server
↓
Capture Thread (30fps)
→
Detect Thread (async)
→
Process Thread (swap+enh)
→
Display Thread (encode)
↓
ONNX Runtime CUDA EP
pyvirtualcam
04 // PERFORMANCE

Speed of Deception

Every millisecond accounted for. Benchmarked on RTX GPU with 1080p webcam input.

Camera Capture (1920x1080)~2ms
Face Detection (RetinaFace ONNX)~3-5ms
Downscale to 480p<1ms
Face Swap (inswapper_128 CUDA)~5ms
Face Enhance (GPEN-BFR-256)~5ms
Upscale to 1080p<1ms
JPEG Encode + WebSocket Send~3ms
TOTAL PIPELINE LATENCY
~20-30ms
30-50 FPS effective • Real-time face swap at full speed
RESOLUTION VS PERFORMANCE
1080p • ~10fps
720p • ~23fps
480p (default) • ~31fps
360p • ~52fps
05 // DOWNLOAD

Ready to Wear
a New Face?

Free. Open source. The mask awaits.

DOWNLOAD FOR WINDOWS
v1.0.0 • Windows 10/11 (x64)
System Requirements
Windows 10+ (64-bit) • NVIDIA GPU with CUDA (RTX recommended) • Python 3.12+ • 8GB+ RAM • Node.js 20+
View on GitHub
Explore the full suite → THE SUITE