Performance does not meet expectations

All of LORA were trained on Pony Diffusion V6 XL.

It's not very stable. The current suggestion is to generate the tail first and then use this LORA.

optional Prompt(but it's unstable)I have discovered a very serious issue where using multiple tail rings can cause image generation to fail.:

• dual tail ring

• 3 tail ring

• thick tail ring

Negative Prompt:

(digit_ring),(horn ring:1.3),(bracelet:1.3),(armband),(ankle bracelet)

Perhaps you can choose their tail rings for your dragons~

I found it difficult for the model to generate tail rings, which is much worse than nose rings. So I trained this LORA.I carefully selected images from e621 for training.But due to the training data, there may be a generation of digit rings, which may be fixed in the future. But I quite like it now~