【Stable-Diffusion】🔰一発でキレイ！ComfyUIの使い方＜STEP3:後編＞ 顔と手の自動修正 #stablediffusion #FaceDetailer #HandRefiner

The video begins with a greeting and an introduction to the topic, which is the continuation of the UI workflow, specifically focusing on the automatic correction of faces and hands in the UI. The presenter mentions that the workflow has become quite complex and advises new viewers to check out the previous videos for a better understanding. The video is structured into three main parts: text input techniques, methods for automatic face and hand correction, and the use of an advanced workflow. Before diving into the face and hand correction, the presenter introduces a convenient custom node for text input to make editing easier and more manageable.

The paragraph delves into various techniques for text input, emphasizing the convenience of using custom nodes. It discusses how to connect text nodes for easier editing and management, the use of preset text for quick input of predefined prompts, and the ability to adjust and combine different text inputs. The presenter also explains how to use the 'WD14 Target' to create prompts from images when no text is available. Techniques for wireless node connections and the use of 'Set Nodes' and 'Get Nodes' for efficient workflow are also covered.

This section introduces the workflow for automatically correcting faces and hands using custom nodes, primarily the 'Face Restore CF' node. The presenter outlines the process of using the 'Face Detailer' and 'Hand Refiner' nodes, along with the 'Impact Pack' for object detection and partial regeneration. The importance of downloading a specific model and adjusting parameters such as denoising, CFG, and the number of steps for optimal results is highlighted. The video also demonstrates how to connect and use 'Set Nodes' and 'Get Nodes' to manage the flow of images and data throughout the workflow.

The paragraph describes an advanced workflow that integrates face and hand correction with image upscaling. It details the process of inputting model and prompt information, generating an initial image, and then passing it through the 'Face Retainer' for facial enhancements. The resulting image is then sent to the 'Hand Refiner' for hand correction. The presenter also discusses the creation of depth maps and masks for precise editing and the use of 'Set Nodes' to direct the flow of images through different stages of the workflow.

The presenter showcases the final results of the workflow, comparing different patterns and techniques used for face and hand correction. They discuss the use of 'Face Restore' and other methods to achieve natural-looking results. The video also includes a comparison of various patterns, such as 'Altimeter SD Upscale' and 'Ultimate SD Upscale,' highlighting the subtle differences in the final output. The presenter encourages viewers to choose the pattern that best suits their preferences.

The paragraph demonstrates how the discussed workflow can be applied to illustrations, showing the step-by-step process of enhancing the image quality through face and hand correction, and upscaling. The presenter also addresses the issue of artifacts that can occur during the upscaling process and provides tips on adjusting parameters like CFG, steps, and noise to mitigate these effects. They emphasize the importance of experimenting with the workflow to find the best approach for a particular image.

The video concludes with a summary of the steps covered in the workflow for upscaling and correcting faces and hands. The presenter expresses the belief that the layered approach to the workflow is effective and encourages viewers to download the workflow from the video description for practice. They also mention that the combined upscale and face detailer workflow, along with the slides, will be distributed exclusively to members. The presenter looks forward to future videos, which will cover the use of animated differences in UI, and invites feedback and questions from viewers.