This person looks undead, but he is not undead.
(But later in the series, you'll see undead monsters that look just like this guy!)
Overcoming the insane V1 model that trained on just a single image, used nanobanana to assemble a reasonable number of training images!
But I guess it's just the curse of being a skull-headed character—teeth really do crumble easily.
Setting the CFG Scale to around 2 or 3 makes it less likely for the face to break up.
Include these prompts at a minimum :
white skin, skull, teeth, green hood, golden necklace, torn robes, green skirt, glowing eyes, yellow eyes, deep eyes hole, long fingernails, no pupils,
For SFW output, it is recommended to set the LORA intensity to 0.8.
NSFW output, it is recommended to set the LORA intensity to 1.0~1.2.
V2 may exhibit strong assertiveness with purple flames, so please adjust using negative prompts (tagging apparently didn't work well).
The image used for this LORA study is only a standing image, so you may want to increase the intensity if you use prompts such as pose specification.
If the intensity is set too high, it forces the painting to become an oil painting, and a purple aura emerges from the entire body.
I thought this LORA would have no effect if he was completely naked, but it seems to inhibit the situation where a woman comes out even though he was prompted to think he was a skeleton-headed man.
The reproduction of the robe is about 70% ......
The necklace appears to be interpreted as a skull shape
Description
I increased the number of training images using nanobanana.