When considering ways to modify ??????? and poses of primarily male characters, I realized that the UFC has a substantial set of images of men categorized by weight in a standard standing pose that could be used to train a LoRA.

It has many knobs (i.e., trigger words) to play with so experiment to get different varieties of men.

Positive Prompt

4fcp0s3 : the trigger word (which can be left off for a subtler effect)

ufc fighter : will solidify the 'style' but will often bring gloves as well

gloves, shorts, bare feet : usually not needed as 'ufc fighter' will often bring these along even when not explicitly called for

flyweight : ~125 lbs or ~57 kg

bantamweight : ~135 lbs or ~61 kg

featherweight : ~145 lbs or ~66 kg

lightweight : ~155 lbs or ~70 kg

welterweight : ~170 lbs or ~77 kg

middleweight : ~185 lb or ~84 kg

light heavyweight : ~205 lbs or ~93 kg

heavyweight : ~265 lbs or ~120 kg

Negative Prompt

gloves, shorts : to help avoid UFC gear when not making a UFC character

shirtless, crop top : to help cover up skin when needed

black background : to help avoid the solid black background from the training data

Dynamic Prompt

{flyweight|bantamweight|featherweight|lightweight|welterweight|middleweight|light heavyweight|heavyweight}

Notes

• With certain prompts, remnants of gloves can generate (e.g., a strip of color or shapes on the hands). To avoid it, adjust the weight of the LoRA or the words in the prompt.

• At times, the LoRA can be quite insistent that the character wear shorts or be shirtless. There is no simple fix except experimenting by changing the words, weights, or models.

• With certain prompts, a black background will generate (as that is how I prepared the training data for the LoRA). To avoid it, decrease the weight of the LoRA and move words around (e.g., 'heavyweight ufc fighter' can be broken up with heavyweight toward the beginning of the prompt and ufc fighter toward the end).

• For full body shots where the faces lose resolution, Hires fix to the rescue.