LoRA for my fursona and main OC Sam Cypris. Sam is a genderfluid/femboy snow leopard. LoRA was trained with multiple poses, a basic outfit of a bell collar, white croptop and skirt, and some variations on markings.

The LoRA isn't as stable as I'd hoped, and the coloration of Sam's spots, ears, and tail-tip are still inconsistent. The trained tags for markings on inconsistent, but generating a heart on Sam's hindquarters sems to work the best.

Both models are a bit on the strong side, so turn them down to 0.4-0.6 if you're generating multiple characters, or are trying to use it to make a similar character.

All version have a tendency to create a snowy background

V1 -

Somewhat more volatile than v2. Tends to generate Sam as a generic or slightly feline character if the species isn't prompted for (snow leopard)

There can be errors around Sam's groin due to the image set containing bulges if you choose to not have one. Trained with more epochs than v2.

Trained trigger words:

SamCypris, white_cat_collar, ButtHeart, HeartSpots, ChestHeart, SamClothes, CrotchHeart

V2 -
More consistent, better tagging of dataset. Higher number of images in dataset with different poses, expressions, and art style than v1, but trained with less epochs.

Better at generating Sam's snow leopard features without prompting for them, but it still generally works out better that way.

Trained trigger words:
SamCypris, SamCollar, ButtHeart, HeartSpots, ChestHeart, SamClothes, CrotchHeart, ShortSnout


Other information:
Tagging words like 'cute', 'androgynous', 'wide hips', 'petite' may change Sam's appearance significantly

Tags you may want to add to get Sam more 'on model':
Cute, freckles, round eyebrows, white cat collar, petite, bulge, wide hips, thick thighs,