>>40893057Not yet. I'm still playing around with it for now. The problem I'm facing currently is that the LoRa seems to work better on some models compared to others. The LoRa was trained on nai base model.
The preview you see above was generated on AOM2-R34-SF1.1 model with these prompts:
>masterpiece, best quality, ultra-detailed, 1girl, dynamic angle, detached crystal wings, heterochromia, pink eye, blue eye, cleavage, bare shouldersBut when I tried the same prompt on one of the holomixes, the LoRa falls apart and the facial features gets a little fucky.
Picrel is done on good_shit. As you can see, its quality is not comparable to the AOM2 model and all of the generations are similar in that it is extremely zoomed in.
