This version fixes the problem of bad hands, It's not an internet celebrity face, the base model integrates WhiteXL_realisticMix, and the training material comes from the frame-to-drawing conversion of video frames. I adjusted the overall model texture using my own refined Lora adjustments. Testing at the size of 512*768*2 produces better results for half-body images, for full-body images, it is recommended to output images at 768*1024*2 to avoid stretching the face. I hardly needed to use face editing tools during testing