使用magicv2增强版的模型训练，推荐大模型使用麦橘v6和麦橘v2增强版等真实模型，使用8400VAE。数据集为100张真人模特的图片，可以将ai风格的原模型真人出图改变为真实感的真人模特，推荐使用20步以上的DPM++ 2M Karras，cfg7，分辨率512x768出图，推荐lora使用在0.6-1范围内，推荐使用质量词语(best quality:1.2),ultra highres,high quality,(realistic, photorealistic:1.4)。可以不用人脸修复插件即可达到很优秀的面部效果，同时兼容多种风格lora，因为使用的主要为半身肖像训练集，建议使用half body之类的半身或者近距离的词汇，防止脸部崩坏。

Train the model using magic_v2 enhanced version, and recommend using real models such as Magic_v6 and magic_v2 enhanced version for large models, using 8400VAE. The dataset consists of 100 images of real models, which can change the AI style original model from a realistic real model. It is recommended to use DPM++2M Karras, cfg7, with a resolution of 512x768 and more than 20 steps to create the image. It is recommended to use Lora within the range of 0.6 to 1, and to use quality words (best quality: 1.2), ultra high res, high quality, (realistic, photosensitive: 1.4). Excellent facial effects can be achieved without the need for facial repair plugins, while also compatible with various styles of Lora. As the main training set used is a half body portrait, it is recommended to use half body or close range vocabulary to prevent facial collapse.