I am very sorry that for real life reasons, the follow-up SD1.5 and SDXL model training for this project has been suspended.
Version v0.3:
Retrained with the base model NAI to get better generalization and tried different base models to see how they worked.
Version v0.2:
Refined the prompts, replaced the base model with anything-v5-Prt-Re, and retrained.
base model: https://civitai.com/models/9409?modelVersionId=30163
Version v0.1:
Of the 189 niji-generated images trained on Lora, 99 percent were boys. Each image is my own carefully selected, fixed the hands and feet of the wrong image. Tag correction is made for each image. It can also be used as a general style of niji, when the weight is increased (0.7 or above) , increase the weight of the prompt accordingly (for example, 1girl, because the higher the weight, the more likely the person is to have a boyish trait such as short hair) .
This is the first Lora Style Model I trained myself, please forgive me if there is any deficiency in it. If you have any suggestions, feel free to contact me ( •̀ ω •́ )✧
Recommended prompt: 1boy,male child
Recommended resolution: 896X512. Then do Hires.fix, there will be better results.
Recommended Embedding: https://civitai.com/models/7808/easynegative
Base model: https://civitai.com/models/9409/or-anything-v5ink