Verified: 3 months ago
SafeTensor
Type | |
Stats | 125 1,183 |
Reviews | (47) |
Published | Nov 19, 2024 |
Base Model | |
Training | Steps: 400 Epochs: 10 |
Usage Tips | Clip Skip: 1 |
Trigger Words | kangaskhan |
Hash | AutoV2 E1D43278D8 |
Created on Civitai
Kangaskhan LoRA trained on Pony upon request. Works very well with a few issues that I'll note below:
Issues:
Eyes can be inconsistent
What helped a ton in my case was actually removing all of my eye quality tags from my positive. Try removing any eye tags to see if it improves clarity
In rear views can add too many anuses
This is due to the training data being primarily semi anthro / feral, so the training data was split 30/70 between having the anus on the tail vs the normal spot
Adding 'tail anus' to my positive helped with this for some inexplicable reason. Adding it to my negative made it worse
Hands can come out stubby and claw-like
Add 'humanoid hands' to your prompt to improve hand consistency