Quick info for people looking: Steps: 40 - 55CFG: 2.5 - 3.5I always loved <a target="_blank" rel="ugc" href="https://venus.chub.ai/characters/2485898">this</a> bot on Chub and despite the lack of images (only five!) I think the v1.0 LoRA turned out pretty okay. But it wasn't enough, so I've spent the last week or so creating and curating new images using a variety of checkpoints to build a new dataset, then threw that back into Civit's trainer to make this v2.0 LoRA for Tenny! It is now MUCH more versatile particularly with backgrounds, no more giant stained glass windows or night-time forests if you don't want them! Seems to work better at lower CFG values between 2.5 and 3.5, with a few of the sample images having been attained with values between 1.6 and 7.5, with higher step counts of 40 - 55 more accurately bringing out the quadruple eye arrangement, which was always an issue owing to most checkpoints' data sets.==== [Deprecated information, v1.0] It will tend to default to an open mouth, but "closed mouth" in the prompt negates this with no issue. Strength/weight: 0.6 (higher will tend to reproduce training images)Sampler: Euler a Steps: 50 CFG: 4

Files with SHA256: 95a9a4f36d02852d4baf87d0de080bf86a657c462fcbd675ccb3c027fc73e657