This is a 3D style LoRA (LyCORIS/LoKr) adapter for NoobAI v-pred 0.5 which aims to make generated 3D images look less janky. B3D should put better lighting, more detail, fluffier fluff and hairier hair on your characters, without compromising on the illustrative aesthetic quality of generated scenes.

Earlier versions of this LoRA are available for FLUX.1-dev and Pony v6.

Note that the base model uses the v-prediction loss function instead of SDXL's standard Epsilon loss. Please follow the instructions here to correctly use v-prediction based SDXL models.