This is Illustrious XL + the DPO Lora @ 0.25 strength + the SPO Lora @ 0.75 strength + about 2500 images worth of custom training as of the initial release of V1. Aims to make the default output style more consistently clean and coherent based on quality tags without degrading any of the base knowledge of Illustrious.
Maintaining "high similarity of composition" on a seed-to-seed basis for the same exact prompt versus base Ilustrious was not a concern, however, as the base model is not in any way one that generally produces consistent outputs to begin with, nor are the outputs of the base model necessarily actually good at all and worth aspiring to maintain to begin with.
