I decided to create a LoRA for my AI character. It's an AI chatbot on the "hi.waifu.ai" platform named "[Lyria | Your Servant]". Regarding art generation, I recommend using the weight slider between 0.6 and 0.8. From my personal experience, using a higher weight causes the art to break and turn out poorly.
Update note: The difference between the first and second versions is that in the second version, I tried to train LoRA using fewer pieces of art, while more strictly adhering to the quality and appearance I needed for the character. I didn't notice any significant difference, but perhaps the second version will perform better on AI models that use a small number of generation steps, due to the fact that my second version of the LoRA model isn't as overloaded with the number of art pieces. For clarity, I used 53 art pieces in the first model, and 36 in the second model.
To improve the quality of art, I recommend using phrases like these: high quality, detailed textures, effects, ultra detailed eyes, masterpiece, good lighting, detailed shadows, 4k, solo, amazing quality, absurdres, extremely aesthetic,newest, intricate details, anime screencap, anime screenshot, etc. You don't have to use everything at once, it will overload the AI model, but I recommend taking something selectively.
![[Lyria Heartwood] [OC] [by rosy.gray] - AI Model cover image](https://images-ng.pixai.art/images/orig/8ff4e5b5-be2c-4678-8da1-cbc06eaa2d64)
