Type | |
Stats | 231 2,269 |
Reviews | (71) |
Published | Nov 16, 2024 |
Base Model | |
Training | Steps: 504 Epochs: 10 |
Usage Tips | Clip Skip: 1 |
Trigger Words | delphox |
Hash | AutoV2 7E5D87701B |
Delphox LoRA trained on Pony. Works fairly well, with a few issues I'll list below. Works quite consistently and is capable of producing some stunning outputs.
Issues:
When genning clothing, the red fur on Delphox's arms can turn into red sleeves and/or attach to other clothing
If this happens to you, I recommend lowering the strength of the LoRA slightly. I usually lowered it to 0.6 from the 0.75 I usually use. Additionally, adding (red sleeves) to your negative can help
Character detail can become inaccurate when genning with lots of clothing
This one's tricky, as the solution is to raise the strength of the LoRA. This can put you in an awkward position where you need to find the exact sweet spot where Delphox's arm fur doesn't become sleeves, but the LoRA is strong enough to enforce the needed character detail
You can try and add to your negative prompt to address these issues. I ran into a few instances of Delphox getting red legs when prompting with a bunch of clothing. Adding (red legs) to my negative helped address this alongside slightly bumping up the LoRA strength
Generation Tips:
Use a max strength of 0.75 to prevent overfit if your model already has Dellphox knowledge (which it likely does)