
This Lora was created using my local kohya instead of the colab, took some time to get all the settings right.
Training based on Anything v3 checkpoint so it works with most anime-like checkpoints.
You only need to type the trigger word 'EXLorelei' and it will generate a fairly accurate image of her. However, the outfit might have some inconsistencies. If you want to make her look more like her version in Pokemon Masters, you will still need to use secondary prompts. Even though I removed all the red-haired samples in the training data, her hair still appears somewhat red. But nothing you prompt gods can't fix.
To recreate her outfit with 100% accuracy, check the prompts and the negatives on the examples shown here. The strength of this Lora is quite high, at 0.8, while anything below 0.3 is too low. I recommend using a value between 0.5 and 0.7.
Don't think I will update this LORA tho, just wanted to test running the whole thing on my pc.
描述:
训练词语: exlorelei
名称: EXLorelei.safetensors
大小 (KB): 147574
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success