From 64e9a355dc17eef9e237f7d3c9c4bde09e5aa6b2 Mon Sep 17 00:00:00 2001 From: vladmandic Date: Wed, 11 Feb 2026 11:15:50 +0100 Subject: [PATCH] update wiki Signed-off-by: vladmandic --- Features.md | 30 +----------------------------- Process.md | 10 ---------- SD-Pipeline-How-it-Works.md | 1 - 3 files changed, 1 insertion(+), 40 deletions(-) diff --git a/Features.md b/Features.md index c5cf5e8..0b0256a 100644 --- a/Features.md +++ b/Features.md @@ -254,35 +254,7 @@ Check the HiDiffusion checkbox in the SD.NEXT webUI in either the Text, Image or

-# Face restore - -## Introduction - -Face restore will try to detect a face or multiple faces in a generated image, then it will do a seperate pass over the face which makes the face have an higher resolution and more detailed. - -## Models - -SD.NEXT has 3 different choices for face restoration: - -- Codeformer -- GFPGAN -- Detailer - -### Codeformer - -CodeFormer, created by sczhou, is a robust face restoration algorithm designed to work with both old photos and AI-generated faces. The underlying technology of CodeFormer is based on a Transformer-based prediction network, which models global composition and context for code prediction. This allows the model to discover natural faces that closely approximate the target faces, even when the inputs are severely degraded. A controllable feature transformation module is also included, which enables a flexible trade-off between fidelity and quality. More here: [Codeformer](https://shangchenzhou.com/projects/CodeFormer/). - -The CodeFormer weight parameter: - -0 = Maximum effect; 1 = Minimum effect. - -### GFPGAN - -GFPGAN stands for "Generative Facial Prior Generative Adversarial Network". It is an artificial intelligence model developed for the purpose of real-world face restoration. The model is designed to repair and enhance faces in photos, particularly useful for restoring old or damaged photographs. GFPGAN leverages generative adversarial networks (GANs), specifically utilizing facial priors encapsulated in a pre-trained face GAN like StyleGAN2, to restore realistic and faithful facial details. More here: [GFPGAN](https://github.com/TencentARC/GFPGAN). - - - -### Detailer +## Detailer Face Hires is a feature that aims to improve the details of faces in generated images. It draws inspiration from the popular Adetailer extension, but simplifies the workflow to a single checkbox, making it easy to enable or disable. diff --git a/Process.md b/Process.md index f0df8b8..84a80a4 100644 --- a/Process.md +++ b/Process.md @@ -12,8 +12,6 @@ This guide covers the process tab in the app. The process tab is where you can u - [Upscale](#upscale) - [Video](#video) -- [GFPGAN](#gpfgan) -- [CodeFormer](#codeformer) - [Remove background](#remove-background) ## Tabs @@ -40,14 +38,6 @@ Upscale your image with the chosen model. Processes a bath or folder of images into a video, gif or png. -### GPFGAN - -Uses the GPFGAN model on your image and applies face restore that enhances facial details. - -### CodeFormer - -Use the CodeFormer model on your image and applies face restore that enhances facial details. - ### Remove background Uses a model to remove the background of your main subject in your image, experiment with the settings for desired result. diff --git a/SD-Pipeline-How-it-Works.md b/SD-Pipeline-How-it-Works.md index 28b0b24..d5fb8f4 100644 --- a/SD-Pipeline-How-it-Works.md +++ b/SD-Pipeline-How-it-Works.md @@ -28,4 +28,3 @@ But this is a short look at the pipeline: From step#3 with cross-attention to blend results 7. Run additional models as needed - Upscale (e.g. ESRGAN) - - Resore Face (e.g. GFPGAN or CodeFormer)