Ye Shunguang Lora from Zenless Zone Zero, specifically designed for the Anima model.
Why do we need yet another Ye Shun Guang model? I tried several Lora models currently uploaded on the site and found their output to be unstable, so I decided to train my own.
This Lora dataset is sourced from Danbooru, using original tags, with prompt structures fully following the Anima template. The dataset contains ~3000 images, supports 1024x1024 and 1536x1536 resolutions (with variable aspect ratios), and has excellent compatibility with different art styles. To use this Lora, you need to use the following prompts:
ye shunguang, white/brown hair, fox ears, fox tail
Ye Shunguang is not a fox girl but I just kept the danbooru tags.
Description
Due to time constraints, the v1.0 version of LoRA was only trained for 6,000 steps. In v1.1, I added another 14,000 training steps, bringing the total to 20,000 steps (6 epoches). This version is more stable and improves the complex details in the official outfits.
FAQ
Comments (1)
The only LORA that works the way anima was designed (using danbooru tags instead of arbitrary keywords, respecting various artist styles, etc).
Props to OP for actually understanding how to train a good LORA. Please do more ZZZ characters!