Restore a face in an old picture: ReActor

<!-- lesson: face-restore -->

Lesson "face-restore", in the group Faces and characters of the lesson list. Steps: 7.

#Consent first

The result is a likeness, not a record of how the person looked. Restore pictures you have the right to use: your own, your family's with their agreement, invented people. Obtain consent before using a real person's likeness, and do not put it into explicit content. See Content rules.

#What you will be able to do

You build a graph that redraws blurred or damaged faces in a picture with a face restoration model.

#What you need

• 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 model of Restore Face 🌌 ReActor: a model, not none.
• Nodes: Load Image, Restore Face 🌌 ReActor, 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, what restoring is (reading)

   A restoration model finds the faces in a picture and draws them again, sharp. It does not know the real face: it guesses the details. The result is a likeness, not a record of how the person looked.

   Restore pictures you have the right to use: your own, your family's with their agreement, invented people. Obtain consent before using a real person's likeness, and do not put it into explicit content.

   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. If the folder models/facerestore_models is empty, the pack itself downloads four restoration models there when ComfyUI starts.

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

2. Load the picture

   Add a Load Image node and choose the picture in its image field.

   If the picture is very small, enlarge it first with an upscale model: the restoration model works on faces of a certain size.

3. Add the restoration

   Add a Restore Face 🌌 ReActor node and connect the picture from Load Image to its image input. In model choose the restoration model; with none nothing happens.

   The usual ones: codeformer-v0.1.0.pth handles badly damaged faces, GFPGANv1.4.pth keeps more of the original, GPEN-BFR-512.onnx is sharp on large faces.

   Files in the old formats (.ckpt, .pt, .pth, .bin) can run code when they are loaded. Take them only from the model's official page.

4. Set how much is changed

   In Restore Face 🌌 ReActor set:

   • facedetection: retinaface_resnet50, the most careful way to find faces;
   • visibility: 1, how strongly the new face covers the old; lower blends them;
   • codeformer_weight: 0.5, for CodeFormer only: closer to 0 is sharper and less alike, closer to 1 is more alike and softer.

5. Save the picture

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

6. When the face looks wrong (reading)

   • The face looks painted on: lower visibility to 0.7.
   • The face is not alike: try GFPGANv1.4.pth or raise codeformer_weight.
   • A face is missed: it is too small or turned away; enlarge the picture first.
   • The rest of the picture stays old: the node restores faces only.

   The facerestore_cf pack does the same with its own nodes and reads the same folder models/facerestore_models.

7. 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.

#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:
  • a Load Image node.
  • image of Load Image: a value is chosen.
• Step 3, Add the restoration:
  • a Restore Face 🌌 ReActor node.
  • IMAGE of Load Image into image of Restore Face 🌌 ReActor.
  • model of Restore Face 🌌 ReActor holds a file and not none (capitals, dots and dashes do not matter).
• Step 4, Set how much is changed:
  • facedetection of Restore Face 🌌 ReActor is retinaface_resnet50.
  • visibility of Restore Face 🌌 ReActor is from 0.5 to 1; Do this for me sets 1.
  • codeformer_weight of Restore Face 🌌 ReActor is from 0.3 to 0.8; Do this for me sets 0.5.
• Step 5, Save the picture:
  • a Save Image or Preview Image node.
  • IMAGE of Restore Face 🌌 ReActor into images of Save Image or Preview Image.

#What next

These lessons go on from this one: Put a face onto a picture: ReActor.

#See also

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