Trained with 9 images, using just the standalone keyword is sometimes enough, but using more specific descriptions yields better results.

Use less weight for custom outfits.

towanatsu, kimono,