A request from an irl friend(they also specifically asked for the last image, I'd rather not know why) :/, turned out surprisingly solid.
trained on Pony Diffusion, strength varies from 0.7-1.0, recommended strength is 1.0, lower accuracy if you're trying to make something specific in term of outfit/view/etc
training done locally with CAME+Rex, 10 epochs, 20 images, 5 and 6 steps for a total of 1020.
trigger word is volskayaOW.
additionnal tags:
mature female, short hair, black hair, blue eyes - helps with accuracy
earrings, white clothes, grey trim, turtleneck, white pants, long sleeves, red socks, black footwear, high heels - her "default" outfit
large breasts - "accurate" size (I don't judge)
negative tags:
buttons - the model has a tendency to turn her top into a suit jacket, this prevents it
double bun - reduces the odds that she gets two hair buns.
frown - most of the data comes from the Sombra animated movie and she's frowning in every single scene.
3d, realistic - optional, changes the shape of the face to be more generic
few things:
first off, a huge thanks to @FallenIncursio for helping make the image of her together with Sombra.
her face is always somewhat in the dark during all of the movie so it has a tendency to add a shadow on it, I have not been able to remove it even with tagging all scenes with dark, shade, shadow, just accept it as a flaw of the Lora(which actually exceeded my expectations).
I actually didn't know about her or Sombra before doing this Lora because I stopped playing Overwatch before Sombra released. Made me realize I had gaslighted myself into thinking OW animations had semi-realistic visuals(they're quite cartoon as per the game AD)
dataset: 2 fanarts(including crop), 0 images from the official concept art(including crops), 0 images from in-game illustrations(including crops), 0 images from in-game screenshots(including crops), 17 images from an animated movie.