Hey !
Arrietty from The Secret World of Arrietty. I tried to get the 3 different outfits.
Trained on WAI illustrious checkpoints.
Training was made with only screenshot from the movie.
v2 :
Some example images are hidden due to evil tag moderation, so please turn off R, X or XXX filter to view them.
Trained on WAI illustrious v16 checkpoints.
Training was made with only screenshot from the movie.
Recommended settings
I recommend a weight of 0.8 for this lora.
The base prompt is :
Arrietty, 1girl, solo, brown hair, black eyes, anime screencap, official style, flat colors,
Customization:
3 outfits :
Red dress:
adventure_dress, red dress, turtleneck, long sleeves
Casual dress:
casual_dress, long beige sleeves, brown top, beige skirt
Pink dress:
travel_dress, pink dress, short sleeves
(optional) boots
2 hairstyles:
Ponytail :
ponytail, clothespin, hair ornament
Loose hair :
medium hair, loose hair
Feel free to give me your feedback ;D
Enjoy ~
1024x1360
Steps 30
CFG 7
Sampler Euler a
Scheduler Normal
v1 :
Recommended settings
I recommend a weight of 0.8 for this lora.
The base prompt is :
masterpiece,best quality,amazing quality, Arrietty, 1girl, solo, brown hair, black eyes, looking at viewer, smile
Customization:
3 outfits :
Red dress:
red dress, turtleneck, long sleeves
Casual dress:
casual dress, long beige sleeves, brown top sleeveless, beige skirt
or
casual dress, long beige sleeves, brown top, beige skirt
Pink dress:
full pink dress, short sleeves
(optional) boots
Hair:
ponytail, hair ornament (optional)
loose hair
Feel free to give me your feedback ;D
Enjoy ~
1024x1360
Steps 30
CFG 7
Sampler Euler
Scheduler Karras
Description
*More accurate
*New data set
*More training
*Trained on the new WAI illustrious v16
FAQ
Comments (4)
Great improvement, will you consider making an anima version?
You read my mind, because i'm currently working on it, learning and trying to train an Anima lora. And i can tell you at first glance Arrietty Anima Lora is impressive ! I really hope the full version of Anima will be released soon. Nevertheless, i'm still experimenting with some parameters. I might release it next week if you are really interested.
@The_Ditruffe great to hear, though I've seen a lot of characters loras from the ones that exist can have an issue with style contamination from the original source, so it might be easier to do a style lora with prompt character directions but I don't know if that's just due to everyone being new
@stygianwizard42 I think it fits pretty well, maybe because i trained it too much, or because i put "ye-pop, deviantart" in the negative prompt and "anime screenshot, official art, flat colors" in the positive one. But the fact that it's a brand new model (and a preview one) partially supported by training software could be the cause of this issue. However, there are so many ways to fail a lora that i can't give you an exact answer, but it seems to work well enough on my side.