Swap your face on a live stream — sharp enough to fool anyone
Faces run through a native high-res channel — measured 4.5× sharper. 720p default, 1080p ultra on demand, auto-downshift under GPU pressure. Dual-face frames, living digital humans that blink and move — all on your own box.
- 25fps live · 1080p ultra · sub-second first frame
- Two faces per frame · third face auto-fallback
- 3-second pre-flight device check
Private deployment · data stays in-house · USDT settlement · verifiable output
Drag the slider — see for yourself
Original vs swapped — see what the native face channel does. Below: a real living digital-human clip.
Hard capabilities, verifiable live
Live swap · native face channel
Native high-res face channel, 4.5× sharper; 1080p ultra averages ≈323ms under concurrency and auto-downshifts to stay smooth.
4.5× clarity · 1080p ≈323ms
HD living digital human
Blinks, head turns, micro-expressions — not a lip-synced still. Cloned face + voice + lip-sync, streamed to WebRTC / OBS.
25fps on a 5090 · ≈0.9s first frame
Image / video swap, refined
Production pipeline: inswapper + GFPGAN / CodeFormer + optical-flow smoothing, batched via a 3-way pool.
GFPGAN 8.8fps / CodeFormer 5.5fps
Live virtual background
Blur / image / green-screen hot-swap, CPU matting with zero VRAM cost — switch mid-stream without a hitch.
≈5ms/frame matting · zero VRAM
Three steps to start
One portrait creates the character
Upload a photo to build the character; optionally restyle hair/makeup before going live.
Go live from your own box
Virtual camera / OBS plug-and-play, with a 3-second pre-flight check before you start.
Stream, platform-agnostic
Output looks like any normal camera; auto-downshift keeps it smooth under load.
FAQ
What GPU do I need?
Entry live swap: RTX 5070 Ti / 4080 16G. Pro (swap + digital human): RTX 4090 24G / 5080. Flagship everything-on: RTX 5090 32G (dual-ready).
Will platforms detect it?
Everything is composited locally and delivered through a virtual camera — the platform just sees a normal camera feed.
Two people in frame?
Yes — left/right slots each bind a target face, and a third face triggers a clean auto-fallback. Solid for interviews and co-streams.
Your face, swapped live
30-minute live demo on your hardware: verify real FPS and clarity with your own material — data never leaves your racks.