Thanks @Rui and @Datou for the valuable insights gained from your workflow.

This workflow is designed to produce chibi style avatars ，it will not  really bring you back to 5 yo .

The main content includes:

1. Using  reference images to guide the style and using Lora to enhance the stylization.

2. Utilizing pulid to reference character features and using instant id to maintain the same posture as the input image, thereby increasing similarity.

3. During image preprocessing, the head is cropped and filled with white to facilitate the production of big-head portraits and to standardize parameters.

4. The Florence model is used to generate prompt words, and the identified "white background" is replaced with "simple background," which may enhance the visual appeal of the output. The prompt "chibi style" is added to reinforce the style.

I test this workflow for many images, it works really well. Enjoy!

致谢@Rui 和 @Datou ，在他们的工作流中学到了有用的知识。

这个工作流是设计来生产盲盒风格头像的，并不是真的可以把你带回到5岁。

主要内容是：

1、使用reference image 来参考风格，使用 lora来增加风格化。

2、使用pulid来参考人物特征，使用instant id来保持和输入图同样姿势，来增加相似度。

3、图片预处理时做了头部的裁切和填充白色图片，方便生产大头照，以及方便统一参数。

4、使用Florence模型来生产提示词，并把识别出的“white background”换成了“simple background”。可能可以让出图更美观一点。添加提示词“chibi style”，用来强化风格。

肝了一整天，测了一大批图，效果还不错，多次抽卡也会有好结果。

目前在线体验地址（仅限国内用户）：

回到 5 岁：https://tt.9syun.com/app/index.php?i=66&c=entry&do=mobile&m=tech_huise&r=index.index.index#/pages/detail/detail?id=1462

checkpoint:

https://civitai.com/models/133005/juggernaut-xl

lora（sliders）：

https://sliders.baulab.info/weights/xl_sliders/rea