Notes on Image Output:
The restored version has good fidelity and generalization, while the generalized version may not be very faithful, but the skin and clothes will be smoother. I sugges* ***ng the restored version.
When refining, some extras were added, and well, those in the know understand.
Because the author is somewhat lazy, most models have no information available. However, there will be some images made using the models below each model. If needed, you can directly look up the information of the image below to draw.
1: Training materials are 512*512, it is recommended that the size of the images produced using this model not exceed 1024*1024. For equivalent results, you must use the base model I used for training! And be sure to use negative prompts!
2: Refer to the cover image for the weights of this lora.
3: ※ Portraits do not need high-resolution restoration, as they will have good results without it. If you want to add details with high-resolution restoration, use the model R-ESRGAN4x+Anime6B with a redraw rate of around 0.3-0.4.
※ If you need full-body portraits, high-resolution restoration is a must, with a redraw rate of around 0.5.
※ For full-body portraits, it is recommended to open ADetailer for the face!
If you like this model, feel free to start running it! Your interactions such as collecting, liking, and returning images are my greatest motivation and support. Thanks again to all supporters!
2. Disclaimer (Declared in 2024):
1. Without permission from the creator (me), do not sell this model on any website.
2. Without explicit written consent, do not use this model to profit on other platforms.