Facefusion: Using Expression Restore with LivePortrait
Facefusion
Demo summary
The video shows the 'expression restore' feature which uses LivePortrait to map eye movements and facial expressions from the target video onto the source face.
Step-by-step
- Enable the Expression Restore feature
- Observe the preview to verify eye movement and facial expression mapping from the target video
- Compare the sharpness of the face with Expression Restore toggled on versus off
Watch out for
- Expression Restore uses LivePortrait which generates a low-resolution image patch
- The face enhancer currently runs before LivePortrait, meaning the final output will appear soft or blurry regardless of enhancement settings
Tips
- Turn off Expression Restore if you prioritize a sharp, high-resolution face over exact expression matching
- Be aware that enabling this feature will increase processing time
Highlights
“the problem with life portrait is that it generates the patch... that is very low resolution and unfortunately I could not get the the face enhancer to run after the live portrait”
All demos from “Face Replacements with FaceFusion 3.0”
- 1:181:28Install FaceFusion 3.0 using PinokioThe video demonstrates how to use the Pinokio browser to search for, download, and install the FaceFusion 3 project, automating dependency and Python environment setup.Pinokio· AI Face Swap Generator
- 3:232:44Basic face swap and face enhancement in FaceFusionThe creator demonstrates uploading a source image and target video into the FaceFusion UI, switching to CUDA for GPU processing, and enabling the 'face enhancer' to sharpen the output.Facefusion· AI Face Swap Generator
- 6:151:47Selecting and downloading AI models in FaceFusionThe demo shows how to switch between different face enhancer models (like GPEN-BFR) and face swapper models, which triggers automatic background downloads in the terminal.Facefusion· AI Face Swap Generator
- 9:471:24Using Expression Restore with LivePortraitCurrentThe video shows the 'expression restore' feature which uses LivePortrait to map eye movements and facial expressions from the target video onto the source face.Facefusion· Video to Video
- 15:411:15Selective face part replacement with Region MasksThe creator demonstrates using the 'region' mask settings to selectively toggle which parts of the face (eyes, mouth, eyebrows) are replaced or kept from the original video.Facefusion· AI Face Swap Video
- 16:561:57Adjusting face mask padding and croppingThe demo shows how to use 'box' mask padding settings to adjust the hairline and edges of the face swap to better blend the replacement with the original video plate.Facefusion· AI Face Swap Video
- 19:041:24Rendering and downloading the final videoThe video walks through setting a frame trim range, clicking 'Start' to process the render through the terminal, and downloading the final MP4 file.Facefusion· AI Animation Generator
- 29:381:34Drop every other frame using FFMPEGAlex demonstrates using a specific FFMPEG command line instruction to remove even-numbered frames from a video to prepare it for temporal consistency fixes.Ffmpeg· AI Video Editor
- 31:281:42Fixing temporal chatter with Topaz Video AIThe video shows importing a half-frame video into Topaz Video AI and using the Chronos model for frame interpolation to create a smoother, more consistent face swap result.Topaz Video AI· AI Video Interpolation
- Watch “Face Replacements with FaceFusion 3.0” →
Video to Video
- 9:471:24Using Expression Restore with LivePortraitCurrentThe video shows the 'expression restore' feature which uses LivePortrait to map eye movements and facial expressions from the target video onto the source face.Alex Villabon