LoHa of Air Shakur (umamusume). Please install extension on WebUI.
After installing extension, put LoHa file in your sd/models/lora folder and write LoRA notation in prompt to applying LoHa. <lora:loha_name:weight>
All uploaded images are generated by AOM3A1
All of my works is free. If you'd like to support me, feel free to buy me a coffee. 👍
Recommended options
LoRA weight around 0.7
Trigger words
Base
air shakur \(umamusume\)
Original Art :
hair ornament, arm tattoo, arm belt, ring, blue sailor collar, blue neckerchief, yellow shirt, sleeveless shirt, bare shoulders, spiked bracelet, studded bracelet, midriff, belt, nail polish, white shorts, short shorts, cutoffs, thigh strap, white footwear
Racing competition costume :
yellow shirt, yellow tank top, side cutout, black choker, necklace, short sleeves, shoulder cutout, crop top, blue collar, cropped jacket, black jacket, open jacket, zipper, wristband, bracelet, belt, black shorts, thigh strap, single thighhigh, blue thighhighs, black footwear
Casual :
choker, hood down, purple hoodie, hooded jacket, long sleeves, sleeves past wrists, off shoulder, purple shirt, purple jacket, (multicolored jacket:1.2), open jacket, black pants, sneakers
I think her clothes is too abstract thing and lack of data, It reproduce not well.
Tracen summer uniform :
tracen school uniform, summer uniform, serafuku, puffy short sleeves, purple bowtie, horseshoe ornament, sailor collar, sailor shirt, purple shirt, white skirt, pleated skirt, frilled skirt, miniskirt, zettai ryouiki, white thighhighs, brown footwear, loafers
Use a model derived from or mixed with animefull model.
Settings
DPM++ Series(SDE Karras, 2M Karras, etc.)
about 20 steps, CFG scale 3.5~6.5
I recommend kl-f8-anime2.vae
CLIP skip = 1
Use Hires.fix to get higher quality image
Upscaler
Latent, Latent (nearest-exact), Latent (bicubic antialiased) or other Latent series.
Denoising strength 0.50~0.65
LoRA training info
trained on sd-scripts by kohya_ss.
Thank you all a lot!
based on Animefull-pruned model
107 images consist of
4x low quality 40 images
8x medium quality 26 images
16x high quality 41 images
=> 1 epoch = 1024 images
replaced character feature tags with air shakur \(umamusume\)
1girl, streaked hair, red eyes, horse ears, hair between eyes, white hair, very long hair, horse girl, brown hair, long hair, green eyes, two-tone hair, simple background, bangs, purple eyes, multicolored hair, horse tail, short hair, brown eyes, grey eyes, yellow eyes, animal ears, orange hair, braided ponytail, single braid, solo, blue eyes
I think these tags are the essentials of Ikuno Dictus.
or unnecessary for training because of duplication.
resolution : 768x768 with aspect ratio bucketing.
other training settings is included in training_data.zip or metadata of LoRA file.
And, special thanks to my best friend. DH.