v1.0 Release!LoRA was trained on, and all images were generated with, Counterfeit v2.5All examples use the seed of "1" with a batch size of 8, meaning no seed cherry picking was done.

Persona 5 Portrait LoRA / LoCon

v1.0Trained on 160 character portraits from Royal, Strikers and Vanilla versions of Persona 5. The dataset was hand picked to have 1 of each character expression and outfit combination, so this release should have a wider variety of those. LoRA strength of somewhere around 0.6 - 0.8 seems to work the best.CFG: 8Sampler: DPM++ SDE KarrasResolution: 512x512

Files with SHA256: 0c79889c791d51043f2cb3da87cdba4c1cd80513755ac29f9ab97b1dce2d14e2