Sign In

Sandshrew

50

291

811

18

Verified:

SafeTensor

Type

LoRA

Stats

291

811

51

Reviews

Published

Jan 21, 2024

Base Model

SD 1.5

Training

Steps: 2,000

Trigger Words

sandshrew

Hash

AutoV2
D080A21EB8

Sandshrew LoRA trained on NovelAI with Kohya training scripts using the DAdaptAdam optimizer at 512x512 resolution. Trained on a low number of source images with inconsistencies, expect issues. Has trouble with tail placement and length sometimes. Can lose some detail when generating clothes due to the lack of clothed source images. Hands come out claw-like, but are compatible with embeddings that enforce a human hand shape. Should have wide compatibility with model mixes based on NovelAI, which is most of them. Trained on a mixed male/female dataset.