dreamoo

I used Redux's excellent style fusion ability and consistency control ability to fuse the dataset and complete the production of Dreamoo's basic dataset. Then, I enlarged the dataset to adjust the resolution, and used flux Lora train in ComfyUI to train the Lora model. The label annotation process was manually performed (because the importance of label annotation accounts for 80% of the Lora model's performance), and finally the Lora model was output. However, because Dreamoo's expression in the dataset is only happy, the model seems to generate happy expressions at any time. Welcome to download and test.

If you want to learn this process and avoid pitfalls in the implementation process, you can follow my YouTube or Bilibili account（ID：啦啦啦的小黄瓜）, and I will update the complete process later.