Type | |
Stats | 624 6,530 2.1k |
Reviews | (166) |
Published | Jan 26, 2025 |
Base Model | |
Training | Steps: 2,805 Epochs: 20 |
Usage Tips | Clip Skip: 1 |
Trigger Words | h4rl3yqu1nn white facepaint, jester cap, jester bodysuit, domino mask, white gloves h4rl3yhy3n4, furry, yellow fur, green sclera, red eyes, sharp teeth, jester cap, domino mask, jester bodysuit, tail, white gloves, muscular red jacket, white shirt, gold medallion, gold earrings, glasses, red lips, blonde hair, long hair, red skirt red shirt, black necktie, black skirt, glasses, red lips, blonde hair, hair bun, white jacket ponytail, white shirt, sleeves rolled up, red lips, bottomless red camisole police uniform, police hat, grey skirt, high heels, black necktie, short hair twintails, white tied shirt, red shorts |
Hash | AutoV2 12BD6F9D46 |
Harley Quinn in her original appearance from Batman the Animated Series. This LoRA includes her traditional red and black jester costume, several other outfits from the show, and her mutated hyena girl appearance from Batman vs. TMNT.
UPDATE 29 January 2025
Added Pony Version
Trimmed the training images of the jester costume to only include front view. Should be able to more consistently get screen accurate color patterns with the Jester costume with this version, but still a little random. I'd say odds of getting screen accurate colors are probably now 30% (up from 10% previously). Tradeoff is that she will likely not have screen accurate jester costume colors in from behind views.
Will incorporate these changes into a future version of the Illustrious model.
Suggested strength: 0.7 - 1.0
There are two "main" triggers:
h4rl3yqu1nn - this is the main trigger. She'll be in her civilian "Harleen Quinzel" form unless you prompt parts from her jester costume. Those triggers are: white facepaint, jester cap, jester bodysuit, domino mask, white gloves
h4rl3yhy3n4 - this is the mutated "hyena girl" appearance from "Batman vs. TMNT." You can reinforce with the following triggers: furry, yellow fur, green sclera, red eyes, sharp teeth, jester cap, domino mask, jester bodysuit, tail, white gloves, muscular
Additional outfits for human form:
Lawyer Harley: red jacket, white shirt, gold medallion, gold earrings, glasses, red lips, blonde hair, long hair, red skirt
Psychologist Dr. Quinzel: red shirt, black necktie, black skirt, glasses, red lips, blonde hair, hair bun, white jacket
Ivy's Hideout Harley: ponytail, white shirt, sleeves rolled up, red lips, bottomless
Rev Your Harley: red camisole
Police Harley: police uniform, police hat, grey skirt, high heels, black necktie, short hair
Shopping Harley: twintails, white tied shirt, red shorts
The traditional red and black jester costume is one of the most difficult challenges I've faced in creating LoRAs. It is asymmetrical, has alternating regions of two tone color, and the parts that should be red and the parts that should be black differ depending on your angle of view. In other words, it embodies everything that AI struggles with. As a result, even with a large dataset it is difficult to reliably get screen accurate output. During testing, I'd say it got the color patterns correct about 10% of the time.
If I make future versions, I'm considering restricting the training images to front view only. It will make it more difficult to put her into some poses, but it may increase the success rate of getting accurate coloration on front view generation.