I've decided to lean into being the single image dataset LoRA guy and make some fun single image LoRAs. Training data is included.
This LoRA will make something really similar to the original image, but this is intentional. Use lower strengths to break out of the original and make something inspired by it, or higher strengths to stick close to the original.
The LoRA was regularized and will respond to the regularization dataset captions as well, most notably "galena" but also (much more weakly) "oatd". Repeating these style words will make their respective effects stronger. You can use higher strengths with these alternate styles.
Long form captioning: santa man, santa suit, fat, arms up, happy, mouth open, chimney, santa in chimney, night, moon, santa hat, white beard, white eyebrows, white moustache, houses, snow, blush, blue eyes, cartoon, bricks, ears
Short form captioning: santa man, arms up, santa in chimney, night, houses, snow, moon
This one effectively has three trigger words: santa man, santa in chimney, arms up
"santa man" is the single strongest caption, it makes a buff-fat version of Santa, and it can be used independently of the rest of the captions if you just want the character.
"santa in chimney" is pretty weak on its own, but chained with "santa man" it should usually give you something akin to the original.
"arms up" is also pretty weak on its own, but is needed to give the arm positioning of the original.
Put any details you don't want from the original into the negative prompt. "white moustache", "white eyebrows", and "white beard" are the most likely to be problems if you're generating a younger and/or female character.