Face Concepts Sets

LoRa of various concepts revolving around face. There is 2 version: A normal version with 10 epoch and a hard trained version with 15 epoch. The result can be quite inconsistant so if you want good result quickly, you can copy the prompts of the image previews.

There is 4 concepts assosiated with 4 trigger words:

• bj-face-pov : pov view of face / best prompts: , veins, saliva, veiny , solo, hetero, upper body, lips

• bj-face- : same that above but with on face and out nostrils / best prompts: , veins, saliva, veiny , (()), out nose, hetero, upper body, lips

• bj-face-pose : suggestion without / best prompts: tongue, solo, hetero, tongue out, hand gesture, oral invation, lips

• bj-face-after : face after / best prompts: solo, ( on face), in mouth, on body, open mouth, (( on )), hetero, upper body, lips

bj-face-pose and bj-face-after can be hard to generate so unfortunately, some tweaks may be necessary.

For bj-face-pov and bj-face-, 40 to 60 steps are needed to have a good-looking image with a high weight on the lora (around 0.9/0.8). For bj-face-pose, the weight needs to be way lower than that (less than 0.6) with the same amount of steps. For bj-face-after, it can be greatly variable, but a weight of 0.9 to 0.7 with 30 steps can be good looking.

The parameters above are to be used with the normally trained version. For the hard trained one, try to lower the weight without touching too much the number of steps.

If you manage to have some great results or if you want to suggest some improvement for the model, you can send me feedbacks in the comment.