Put a face onto a picture: ReActor

<!-- lesson: face-swap -->

Lesson "face-swap", in the group Faces and characters of the lesson list. Steps: 9.

#Consent first

Only someone who agrees may be shown this way: yourself, a person who gave permission, or an invented character. Obtain consent before using a real person's likeness and permission for the source material, and follow local law. Do not put a real person's likeness into explicit content; exclude anyone who is or appears under 18; do not pass a person off as saying or doing what they did not. See Content rules.

#What you will be able to do

You build a graph that puts a face from one picture onto a person in another picture you already have.

#What you need

• It goes on from Restore a face in an old picture: ReActor. The card says so; it is a hint, not a lock.
• Node packs: ComfyUI-ReActor. You install them in ComfyUI yourself, then choose Sync ▸ Sync now. Until then the card says the lesson does not fit and names the pack.
• Model files the lesson checks:
  • for face_restore_model of ReActor 🌌 Fast Face Swap: a model, not none.
• Nodes: Load Image, ReActor 🌌 Fast Face Swap, Save Image / Preview Image.
• The app downloads and installs nothing. The step texts below say which files and folders, and which ComfyUI template names them.

#The steps

The words of each step, as the lesson shows them. Node names are in bold.

1. First, about faces (reading)

   Only someone who agrees may be shown this way: yourself, a person who gave permission, or an invented character. Obtain consent before using a real person's likeness and permission for the source material, and follow local law. Do not put a real person's likeness into explicit content; exclude anyone who is or appears under 18; do not pass a person off as saying or doing what they did not. This lesson teaches technique.

   ReActor finds a face in one picture and puts it onto a face in another. The example: a character you generated gets the face from your own photo, or your own face goes into a picture you drew.

   The nodes of this lesson come from the ComfyUI-ReActor pack. A pack runs other people's code on your computer: install it yourself through ComfyUI's manager, and only a pack you trust. The pack has its own detector that refuses explicit pictures.

   The lesson builds the graph from an empty canvas: "File ▸ New project".

2. Load the picture to change

   Add a Load Image node and choose the picture whose face is replaced in its image field.

3. Load the face

   Add a Load Image node and choose the picture with the face in its image field. It must be another file than the first.

   A frontal, sharp face works best. If either picture has several faces, the node takes them in order from the left; its fields input_faces_index and source_faces_index choose which.

4. Add the swap

   Add a ReActor 🌌 Fast Face Swap node. Connect the picture from Load Image to its input_image input and the face from Load Image to its source_image input.

   In swap_model choose inswapper_128.onnx; the pack downloads it to models/insightface when it is installed. Set facedetection to retinaface_resnet50, the most careful way to find faces.

5. Sharpen the new face

   The swap model works at a small size, so the new face comes out soft. In ReActor 🌌 Fast Face Swap choose a restoration model in face_restore_model, for example codeformer-v0.1.0.pth, set face_restore_visibility to 1 and codeformer_weight to 0.5.

   These do what the lesson "Restore a face in an old picture: ReActor" explains.

6. Save the result

   Add a Save Image node and connect the SWAPPED_IMAGE output of ReActor 🌌 Fast Face Swap to its images input.

7. Swap, adapter or LoRA (reading)

   A swap changes a picture that already exists: the pose, the light and the clothes stay, only the face is new. An adapter such as IP-Adapter FaceID draws a new picture around the face. A LoRA trained on the character holds face, body and clothes in every picture.

   The built-in knowledge "Identity: adapter, LoRA or swap" compares them.

8. When the result is bad (reading)

   • The face looks pasted on: lower face_restore_visibility to 0.7, or try GFPGANv1.4.pth.
   • The wrong face is changed: set input_faces_index.
   • The skin tone differs: pick a source photo with similar light.
   • Nothing happens: no face was found; use a larger, frontal face.

   A face that is small or broken in a picture you drew is repaired in "Fixing faces and hands with a detailer".

   The InsightFace models the swap uses are for non-commercial use only.

9. The graph is ready (reading)

   Save the graph with "File ▸ Save workflow for ComfyUI…", open the file in ComfyUI and run it there. This app runs nothing in ComfyUI itself.

   Nothing made this way may pass a person off as saying or doing what they did not.

#What the lesson checks

A reading step is done when you press Got it, next. A doing step ticks itself off when the canvas holds:

• Step 2, Load the picture to change:
  • a Load Image node.
  • image of Load Image (picture to change): a value is chosen.
• Step 3, Load the face:
  • a Load Image node.
  • image of Load Image (face) is not the same file as image of Load Image (picture to change).
• Step 4, Add the swap:
  • a ReActor 🌌 Fast Face Swap node.
  • IMAGE of Load Image (picture to change) into input_image of ReActor 🌌 Fast Face Swap.
  • IMAGE of Load Image (face) into source_image of ReActor 🌌 Fast Face Swap.
  • swap_model of ReActor 🌌 Fast Face Swap: a value is chosen.
  • facedetection of ReActor 🌌 Fast Face Swap is retinaface_resnet50.
• Step 5, Sharpen the new face:
  • face_restore_model of ReActor 🌌 Fast Face Swap holds a file and not none (capitals, dots and dashes do not matter).
  • face_restore_visibility of ReActor 🌌 Fast Face Swap is from 0.5 to 1; Do this for me sets 1.
  • codeformer_weight of ReActor 🌌 Fast Face Swap is from 0.3 to 0.8; Do this for me sets 0.5.
• Step 6, Save the result:
  • a Save Image or Preview Image node.
  • SWAPPED_IMAGE of ReActor 🌌 Fast Face Swap into images of Save Image or Preview Image.

#See also

• Node packs
• Unsafe model formats
• Content rules
• How lessons work