About this version

This is the next model evolution cooked with slightly different approach to training and mixing, with the main focus on realism (at some cost in artistic elements like painting, drawing, cartoons...). I took a few steps back at that point, retrained some "questionable" parts and added a few LoRAs.

Noticeable changes:

Much better prompt adherence

Overall image quality and balance is a bit better

Way less hallucinations, random backgrounds or non-prompted details

* Occasional unstable images can appear, with excellent framing and quality, but some bad elements or details > this can be resolved with simple step number adjustment (a few more or less steps will usually solve the problem)