Second attempt, as first one didn´t generate, not sure why. First step on the development of a AI character called Eider, only intended as an intermediary step at a more distilled final version. It is not based on any real person and it is not trained on NSFW images.

It seems to work best with Jib MIx, Juggernaut and Analog Madness. These models appear to produce better results on draft mode at CFG 6 and low step count (10-12).

I have been keeping the LoRa at 0.65 strenght as over that it overpowers the image in a negative way. this works fine for me as an intermediary step towards v3. It seems to work best for 1/3 front facing body portraits.

The LoRa is consistent enough if kept in the same composition (close portrait, 1/2 body portrait, full body...) and same model (Jib Mix, Analog...). A full body in Analog Madness will produce a substantially different result to a close portrait with Jib Mix. But we are getting there...