<p>Asian Girl Face:Training based on synthetic data.</p><p>Yes, perhaps you may think that this character is very similar to some real people in reality, but please believe me, it can only be a coincidence. The model training method I use is not a very special secret. At the beginning, I will have a training data set based on real people. There will be some pictures of real people classified according to face shapes. Then, through character consistency algorithms such as ipadpter, instantID, etc., a batch of new AI-generated characters that are equally beautiful, have some similarities but are actually significantly different are generated by mixing facial features of multiple characters. After accumulating a certain amount of materials, I will obtain some LoRA models of characters in the first stage. Then, in the second stage, in an AIGC environment similar to ComfyUI, I will randomly generate high-definition pictures with real photo texture and different occupations, clothing, expressions, postures, and environmental backgrounds by using nodes similar to SDXL Auto Prompter, etc., obtaining a larger and richer training set that is almost indistinguishable from real person photo materials. Well, after a new round of training, we have obtained a brand new character model. She is so beautiful, real, and has a rich and diverse life experience. But as you and I both know, she is a virtual character. You can use her in some research-oriented AIGC generation tasks, and you can also explore commercial uses. As long as you do not violate relevant laws and rules regarding content security and morality, there is no need to worry too much about unnecessary disputes caused by facial similarity.</p>

Files with SHA256: 49653037ec6d4b49581242070f0494f958d060dd5b53786366f29974784d31a9