Get Started
The Lethal Stack

Mission 3 of 4

The Lethal Stack, Module 3: FaceFusion, one photo onto any video

FaceFusion takes one photo of a face and puts it onto the person in a video, frame by frame. In this module you install it, open the app, run your first face swap, then learn the settings that make the result look real and the ones that make it fast.

60 minIntermediateFreeStep 1 free

What you'll do

FaceFusion installed in its own clean environment, with FFmpeg and Git
The app open in your browser, and a tour of every panel that matters
Your first face swap, on a photo and then on a video
  • The settings for a realistic result: the right swapper model, the face enhancer and a good source photo
  • One command that runs a swap with no app at all, for batches and overnight jobs
Before you start
  • A Windows, Mac or Linux computer with at least 8 GB of RAM and about 5 GB of free disk space
  • A graphics card is optional. Without one it works, slowly. We give you our real timings
  • A photo of a face you're allowed to use, and a short video to put it on
  • About an hour. Every command is given to you to copy
Your progress0/10
01/10What FaceFusion does, and what a real result looks like0 done
STEP 01 of 10~4 min

What FaceFusion does, and what a real result looks like

See a real output and the honest speed before you spend an hour installing.

FaceFusion ↗ (over 30,000 stars) calls itself an "industry leading face manipulation platform". The headline feature is the face swap: you give it a source (one photo of a face) and a target (a photo or video of someone else), and it redraws the target with the source's face, keeping the target's head movement, lighting and expression.

Our own run. Left: the source photo. Middle: one frame of the target video. Right: the same frame from FaceFusion's output. Both people are AI generated, nobody real.

Look at what changed and what didn't. The eyes, eyebrows, nose, beard and skin are from the photo. The hair, the head shape, the clothes and the room are from the video. That is how every face swap works: it replaces the face, not the head.

It is more than a face swapper

The app is built from processors, tools you switch on one at a time. The ones in version 3.9.2 are:

  • face_swapper: puts the source face onto the target. The main event.
  • face_enhancer: sharpens and repairs a face. Turn it on with the swapper for a cleaner result.
  • lip_syncer: makes a face's mouth follow an audio file.
  • age_modifier: makes a face look older or younger.
  • expression_restorer: brings back the target's original expression if the swap flattened it.
  • frame_enhancer: upscales the whole frame, not just the face.
  • background_remover, frame_colorizer, face_editor, deep_swapper and face_debugger: cut out backgrounds, colour black and white footage, adjust features, use ready-made face models and see what the detector sees.

The honest speed

Everything runs on your own computer, so speed depends on your hardware. On our laptop with no graphics card, a 5 second clip (121 frames, 1080 by 1920) with the swapper and the enhancer both on took 12 minutes 27 seconds: about 6 seconds a frame. A PC with a modern NVIDIA card does the same job many times faster. Step 8 shows you how to get more speed.

Warning
Only use a face you own or have clear permission to use, and never to pass yourself off as a real person. FaceFusion includes a filter that blocks explicit content, and its licence forbids harmful use. Module 1 has the full ground rules.
Checkpoint
You know what a source and a target are, what a processor is, and roughly how long a clip will take on your computer.
Stuck? Ask the community
Next mission · The Lethal StackThe Lethal Stack, Module 4: A real-time voice changer on your own PC
→