Type | |
Stats | 155 1,044 |
Reviews | (34) |
Published | Apr 30, 2024 |
Base Model | |
Usage Tips | Clip Skip: 2 |
Trigger Words | smearglepkmn |
Hash | AutoV2 6862B61085 |
(Trained at 768px with 64 images for ~6400 steps using Prodigy optimizer on terminal SNR FluffyRock e103 for 1.5; Similar settings at 1024px on PDXL.)
I've desired to create this LoRA for a while now, if my profile image hints. I think it performs well enough, at least compared to generating without it. My only issue is the uneven amounts of arm/leg marks, but at least is consistent with itself. (Majority of generations will have 3 on both arms, 1 on both legs, for example.)
I hope you enjoy it regardless. I have tried to deliver a PDXL version at the same time as well, but I still lack hardware and knowledge to generate/train on it. Please excuse if it is not satisfactory.
Activation keyword is smearglepkmn; LoRA works fine at weight 1, but works as well as low as 0.8.
Recommended positive keywords: tan fur, eye markings, noseless, thin tail, long tail, tail tuft
Optional keywords: [color] tail tip (The LoRA receives prompts for the tail's brush color very well from tests. Common colors with the tail tip keyword will often work with little issue. Dataset is majority green, so it is the "default" so to speak.
Recommended negative keywords: black sclera, digitigrade