Upload a facial reference image, a facial position reference image, and an IPA style reference image (this last one is optional, as the IPA style reference can use the same image as the facial style reference, and users can switch the reference image source for the IPA module themselves). This allows the generation of character images that closely match the facial features of the reference photo, in a specific style and pose.

The process defaults to starting from an empty latent image to generate results but also supports image-to-image generation based on the facial reference photo (users need to switch this option themselves). If users choose the latter option, they need to adjust the denoising parameter from 1 to 0.65 at the sampling node.

上传一张面部参考图，一张面部位置参考图，一张ipa风格参考图（也可以不上传此图，ipa风格参考图可以共用面部风格参考图，用户可以自行切换IPA模块的参考图来源），可以生成与面部参考图容貌特征非常接近的，特定风格，特定姿势的人物图片。

流程默认从空的latent图像开始生成结果，也支持基于面部参考图进行图生图。用户如果选择后者，那么需要在采样节点，把降噪参数从1改为0.65。