(Trained at 768px with 27 images for ~3800 steps using Prodigy optimizer on terminal SNR FluffyRock e103; Same parameters at 1024px for ~1600 steps on PDXL.)
Another experimental LoRA, but possibly more unstable. Dataset is almost entirely screencaps from titular anime. However, regularization was used to avoid style of the source baking into generating.
Unfortunately, some aspects may be inconsistent, and default "sitting" pose is very prominent, largely when "front view" or "from front" are used for angle. Please excuse it.
Activation keyword is wistkiki; Suggested LoRA weight is 1. Lower than 0.85 may yield undesirable results.
Recommended positive keywords: feral, quadruped, white eyelashes, black fur, pigtails, low twintails, pink eyes, white nose
Recommended negative keywords: black eyelashes, pink nose, mouthless (Mouth may not appear or be hard to perceive, largely for 1.5 model. This is a result of some lower quality screenshots, where the mouth is not clear. Can also be alleviated if prompted for specific mouth shapes, like smiles or frowns.)
Version trained on NoobAI, mostly tested for zatochu's merges, such as "Zoinks" and "Jinkies." Dataset and training are mostly similar of the Pony Diffusion version, so prompts are similar.
(Preview imagery generated in wildcard batches, for time conservation. Clean-up and upscaling are exempt for this reason. Prompts may contain nonsense as a result, except for character details.)
1. 재게시된 모델의 권리는 원 제작자에게 있습니다.
2. 모델 원작자가 모델을 인증받으려면 공식 채널을 통해 SeaArt.AI 직원에게 문의하세요. 저희는 모든 창작자의 권리를 보호하기 위해 노력합니다. 인증하러 이동
