A lora model of my favorite version of Harley Quinn from Batman Arkham Knight, trained over 94 images for 1024 using AAA as a training model
Trigger word: harleyqbak
Weight: 1 to 0.4 (0.7 is recommended)
Looks tags: short twintails, multicolored hair, blonde hair, red hair, blue hair, blue eyes, pale skin, makeup, eyeshadow, lipstick,
Clothes tags: asymmetrical clothes, multicolored clothes, choker, o-ring, cleavage, puffy short sleeves, armlet, red vambrace, single vambrace, black elbow sleeve, single elbow sleeve, corset, frilled skirt, asymmetrical legwear, pantyhose, cross-laced footwear, asymmetrical footwear, boots
Additional information:
Webui I use
AAA Anime VAE SDXL v1_4 as VAE
Sampling method: Euler a
Schedule type I set it to Karras
Sampling steps I go 20-30
4xNomos8kDAT as an Upscaler
Hires steps I go 8-15 (I don't always use hires fix)
Denoising strength I go 0.4
Upscale by 1.2 to 1.5 depending on the starting image dimensions
Detail Daemon at default settings
Imagine dimensions vary depending on what I'm trying to do
I use 5 CFG Scale and CFG Scale Fix set at 9
uddetailer for faces, I use mediapipe_face_mesh set Dilation to 50 and Y offset to 30 to get the full face