Cracked skin is another concept which the AI doesn't seem to be able to understand very well, so I have trained a LORA to take care of this.

It takes a bit to get it working proficiently using txt2img, but once you've got it down it works like a charm. Does especially well with offset loras/models. After testing for a while, I ended up achieving my best results using all of the trained tags plus an offset lora and the cracked skin lora both set at 0.6.

It is capable of both cracked skin and glowing cracked skin. Cracked skin can be achieved simply prompting those words, but for the latter though, I highly advise using an offset lora or prompt weighting "dark skin, colored skin", otherwise you end up struggling to get visible results. You may possibly have to use all of the tokens together, but I'm not entirely sure on that. You can also easily inpaint with it, too, which I would recommend over using txt2img if you care about precision.

Eventually I want to do a lora for veins as well, but apparently building datasets for this kind of concept is fairly difficult.