My first try at training on manga crops. Didn't even bothered to crop the speech bubbles away and it still worked just fine though I played safe and trained with low learning rate.

This LoRA was trained on the main cast of a manga series that goes by the absurdly huge name of: 'Eroi Koto Suru Tame ni Kyonyuu Bishoujo Dorei o Katta Hazu ga, Oshishou-sama to Shitawa Rete Omotta Toori ni Ikanaku Naru Hanashi'.

It aims to learn the 4 'main characters' plus each individual part of each of their main outfits. Which is quite a challenge since they are multi-racial with different skin colors, ear shapes, and intricate outfits.

Dataset: ~98% monochrome manga crops ; ~1% monochrome manga covers ; ~1% colored manga cover crops

Negative Triggerwords: monochrome, greyscale, speech bubble, text

Main character Triggerwords ( sometimes you get better results without '\(eroi hanashi\)' ):

• halohalolinne \(eroi hanashi\), blue green eyes, very long silver white hair, elf ears

• filia \(eroi hanashi\), blue eyes, light brown hair

• shina \(eroi hanashi\), fire red eyes, long red brown hair, red brown shina cat tail, black ribbon on animal ears

• amor \(eroi hanashi\), magenta pink eyes, pink hair, dark skin, pointy ears

Outfits:

• halo hat, halo magician robe, halo magician dress, white gloves, halo magician boots, white thighhighs

• halo loungewear

• halo pajamas

• filia black slave collar, filia sleeveless shirt, filia white skirt, filia large brown gloves, filia loafers and socks, filia black thigh strap on her left leg

• shina jacket, shina strapless bra, shina unbuttoned shorts with black straps, shina cowboy boots

• amor long sleeved tunic dress, white shoes