An extensive series of painstaking refinements between the previous version and a variety of other models I've published (as well as some prototypes I haven't), including some Pony models, thanks to my discovery of a means to merge the two architectures together. This is about forty merges down the line from the launch version, and is a fairly noticeable departure from it, despite keeping and refining the general aesthetic, you may find your prompts and seeds to have rather different results with this version. It's able to use Pony (score_9 etc.) prompting better than the prior version, but benefits from a slightly higher CFG scale (whereas I had been sticking with 4, this one seems to like 5 the most.