Replace selected people while the video plays. Bind each role to a different face, keep everyone else unchanged, and process the frame on your own device.
Research prototype · Selected-device validation · Partnership enquiries
Interactive mapping illustration, not rendered face-swap footage. Try the toggles or switch positions.
Two native prototypes. Shared role-matching concepts, with capture and display tailored to each operating system.
Select a video window and manage replacements from a compact floating toolbar. Collapse it to a circle while the source player remains usable.
Share a selected app through Android’s system screen-capture flow. A floating role panel manages references without returning to the launcher.
A person can move, leave the scene, return, or appear in a photograph. Every visible instance needs its own identity confirmation.
Choose a detected face from the original video. This becomes the identity reference for the role.
Assign a different replacement image to each role. The replacement photo is not used to identify the original person.
Add other poses of the same person, inspect matching scores, and remove individual samples. References and thresholds persist.
A fast model does not automatically mean smooth screen playback. We keep engine timing, actual output rate, capture, and display separate.
See measurements and limitations ↗450 frames paced at 30 fps; 3 processing frames exceeded 33.33 ms. These are engine timings on a repeated still image, not an end-to-end moving-video guarantee.
The current face-replacement engines run as native Mac and Android apps. This website introduces the technology; it does not upload or process your video.
Yes. Roles A and B can use different replacement photos. Binding uses identity references, rather than left-to-right face order. Unbound people remain unchanged.
Each visible instance is tracked and confirmed independently. The same role can then use the same replacement face for both instances, including a face visible in a photograph.
No. The prototype has selected-fixture measurements, while robust end-to-end performance under motion, cuts, occlusion and large head rotations remains under evaluation.
This is a technical prototype for integration discussions. The current model bundle includes research-restricted pretrained models; commercial distribution requires appropriately licensed models and a separate integration scope.
Tell us the target device, number of faces, video source and deployment needs. We can discuss technical evaluation, custom integration and distribution partnerships.