This model is uploaded just for a test. A better LoRA of amiya is here: https://civarchive.com/models/121403/amiya-arknights

All pts can be found on https://drive.google.com/drive/folders/1eFmfDmp2yalPXj0K1gEQBCMGvusqW9eP?usp=sharing

This is the README on the abovementioned link:

User Guide for This Embedding

The current path contains the following:
1. 'README.txt', this document, a guide for usage.
2. 'embedding' folder, containing the intermediate embedding file of each step in the training process (named 'amiya-yyy.pt', where 'amiya' is the role name, yyy is the number of training steps, such as 13250, and the full name after combination is 'amiya-13250.pt')
3. 'xxx.png/xxx_meta.txt', is the test images of this embedding, and the detailed prompt and model configuration used. 'xxx.png' contains a table composed of pictures. The integer number on the top represents the number of steps, while the float number on the left represents the weight used in prompt. The table reflects the actual performance of embedding in various states.

Usage:
1. When using, copy the 'amiya-*.pt' files under the 'embedding' folder to the './embeddings' sub folder of your webui working path.
2. Check 'xxx.png' to find out the best state range of embedding. Specifically, please pay attention to whether the following conditions appear in the image:
  a) Bad background, and the characters are distorted or pixelated (too high weight)
  b) Whether the character's appearance features are seriously lost (too low weight, or too low steps)
  c) For the pictures of bikini and nude, whether the characters still wear the original clothes instead of swimwear/stripping (the number of steps is too high, or the characters are coupled with his/her clothes)
3. Based on the basic configuration of 'xxx_meta.txt', use the X/Y Plot function to further tune the steps and weights
4. Configure your own prompt, draw your image.

[!!!IMPORTANT!!!] Instructions:
1. This embedding is trained with a self-modified version of stable-diffusion-webui, which uses program script to automatically grab the images of xxx from the Internet, and conducts training after image-recognition-based data cleaning and tagging. The developer of the program script is [email protected]. There is no human involved in the whole training process. Therefore, the quality of this embedding will largely depend on the quantity and quality of data on the Internet.
2. Our 'xxx.png'/'xxx_meta.txt' is only a simple test and cannot fully represent the real quality of embedding. The quality of the final output image depends on further prompt and settings.
3. In 'xxx.png', the used keywords of character features are automatically recognized based on images from the Internet. Therefore, there may be some errors and omissions, which may lead to deviations in the characters in 'xxx.png' (e.g. hair color, long/short hair, chest circumference, clothes style, etc.). these problems can be easily fixed by adding or deleting certain keywords.
4. For the current character 'amiya', the feature keywords detected by the program script specifically include: [brown hair, rabbit ears, animal ears, long hair, blue eyes, jacket, jewelry, black jacket, hair between eyes]