Search

6.4. Detailer

Chapter
6.Feature

Learning Goals

When generating images with AI, it's common to encounter issues like distorted faces or incorrect hands. While perfect solutions are challenging, tools like Hand Fixer and Face Detailer can address a significant range of these problems. Let's study these tools:Hand Fixer/Face Detailer
Plain Text
복사

Hand Fixer

6.4.1.HandFixer(ControlNet)

Using ControlNet's LineArt can help improve hand generation in images.
Since image generation AI often struggles with rendering hands perfectly, providing appropriate assistance with tools like LineArt can help the AI draw hands more accurately.
Generating such images entirely from scratch with AI can be very challenging. However, by using the above image with ControlNet, you can easily create various images while constraining the structure of the hands.
In the basic workflow of 4.2.1. ControlNet (SDXL), this workflow applies the MistoLine model and preprocessor, which follows lines well and draws accordingly. The application method can vary slightly depending on the type of ControlNet. Using ControlNet, you can create various images while maintaining the structure of the hands.

6.4.2.HandFixer(Graphormer)

You can generate hands again using ControlNet's Graphormer.
The workflow involves generating an image and then using the MeshGraphomer Hand Refiner node to correct the hand shapes. Following that, use ControlNet Depth to regenerate only the hand portions.

Face Detailer

6.4.3.FaceDetailer(ImpactPack)

You can regenerate the face using FaceDetailer.
The workflow involves generating an image and then using the FaceDetailer node to detect the facial area. After that, regenerate only the face portion.

6.4.4.EyeDetailer(ImpactPack)

Using MediaPipe and FaceDetailer, you can detect different parts of the face and then regenerate them accordingly.
The workflow involves generating an image and then using the MediaPipeFaceMesh Detector Provider node to detect the eye region. Following this, use the FaceDetailer node to regenerate only the eye area. Additionally, you can use other options in the MediaPipeFaceMesh Detector Provider node (such as face, mouth, eyebrows, eyes, and pupils) to detect and regenerate other facial parts besides the eyes.