Player not loading? Watch on YouTube
This tutorial explains a local AI face swap workflow using FaceFusion 3.8 and Pinokio. The presenter demonstrates installation on Windows and points to Pinokio downloads for Mac and Linux. The setup includes activating a local server, accepting prompts to install Anaconda and Python components, and opening the FaceFusion interface in a browser through its URL. The presenter also shows a QR code for accessing the server from another device.
Footage preparation comes first. The presenter trims the target video in Premiere Pro to keep a close view of the face, explaining that the version used struggles with varied face movements. When the face disappears, the edit returns to the original footage. For the source image, the workflow combines a screenshot from the video with a photo of the presenter in Google Flow. The presenter corrects an unwanted red tint in the generated hair before downloading the image. This image preparation uses a separate service, so the demonstrated workflow is not entirely offline.
In FaceFusion, the presenter recommends HyperSwap 1A at 512 × 512 resolution with a face swap weight of 0.85. Other settings stay at their defaults, though FPS can be adjusted. The final steps select the reference face in the preview, place the generated image in the source field, and use the trimmed video as the target.
