【沼すぎる～】AI画像の手の修正の方法（レベル１）【stable diffusion】

Alice from AI's in Wonderland introduces the topic of hand corrections in AI art, a common challenge for artists. She explains her approach to correcting hands in both live-action and animation, emphasizing the use of Photoshop. Alice shares her experience and the trial-and-error process that led to her current method. She also discusses the limitations of using Style Diffusion without paint software and the challenges of creating realistic hands. The video begins with level 1 hand corrections using an animated image, highlighting the ease of making fill-ins in animation compared to live-action. Alice mentions the Depth Map Library & Poser extension, which is no longer updated and may not be the best choice for new projects. Instead, she suggests using only the hand depth maps and explores alternative methods for hand correction.

Alice discusses the availability of hand depth maps created by Daltwa TV, which are freely downloadable and can be used to improve hand corrections. She apologizes for using a screenshot without permission and encourages viewers to follow Daltwa TV. The focus returns to hand correction, with Alice explaining the use of inpainting and control nets for level 1 hand adjustments. She guides the audience through the process of using Canvas Zoom for fullscreen painting and selecting the 'Inpaint' control type. Alice also introduces the use of negative embedding, specifically BadHand5, to improve the hand's appearance. The video demonstrates the inpainting process and the resulting image, noting some minor issues that could be further refined using Photoshop.

Alice delves into advanced Photoshop techniques for fine-tuning hand details, starting with a new Photoshop project and importing the previous image. She discusses the challenge of adding a thumb from scratch and instead uses a trick to borrow the thumb from the other hand. Alice explains her selection process using the Magnetic Lasso Tool and the Select and Mask tool for precise adjustments. She emphasizes the importance of practicing hand correction in Photoshop and shares her learning journey, including her inspiration from MappyPhoto's YouTube videos. The video showcases the process of transplanting the selected thumb, flipping, and adjusting it to fit the right hand. Alice also covers the use of layer masks for blending and making adjustments with painting tools, highlighting the importance of layer management for easy modifications.

Alice continues to discuss hand modification, focusing on minor adjustments and the use of brushes for painting over the image. She emphasizes the flexibility of using layers for easy modifications and the importance of blending in the corrections naturally. Alice notes the differences between working with anime-style and realistic images, mentioning the use of tools like the Spot Healing Brush or the Clone Stamp for realistic hand corrections. She shares her personal goal of seamlessly modifying AI-generated images and encourages viewers to experiment with different hand correction methods. Alice also mentions her use of depth maps from DawatreTV for the LoRA corner and introduces the concept of using LoRA for hand depiction, despite the potential challenges in content moderation.

In the LoRA corner, Alice discusses the use of LoRA for hand depiction, specifically for creating images of holding panties, which is a daring theme by YouTube standards. She acknowledges the potential risks but is motivated by the desire to feature content related to hand depiction. Alice explains the process of generating images with LoRA, the use of tags, and the training data behind the AI model. She shares the generated images, noting the influence of manga or illustration images in the results. Alice selects an image with moderate hand damage for further correction using a hand tremor map provided by Daltwa TV. She demonstrates the inpainting process using Photoshop and Control Net, resulting in a well-corrected hand. Alice concludes the video by encouraging viewers to try LoRA and promises to introduce methods for hand correction using only stable diffusion web UI in future videos.