V3 is a refined and polished version. This is probably the final version of this model. Overall, the recommendations are the same.
V2 is stable. Fixed an error when noise instead of image was created
This model is close to realism. Something a little more semi-realistic. Use any SDXL resolution you like.
Use Sampler - Euler a or DPM ++ 2M SDE and Karras
I recommend the use 832 x1216 -- 960 x 1520 resolution to create portrait images
Steps - 25+
CFG scale - 5