Hi there !
This is my first LoRA. I've been trying several things before releasing it (V1 is actually the 6th version). I'm not satisfied yet, many things can interfere with what I prompt for, so please feel free to give criticism about it.
V2 (Illustrious) : trained with more steps with the same (sometimes distorted) dataset as Pony version. It sticks better to the original style, with more expressive facial expressions. Also, the model is more able to recognize side characters (Sacchan, Nidad, Alienman). Nothing really changed with the prompt. You may want to add "green eyes" in negative prompt, as it happens sometimes.
V1 (Pony) : Faces and bodies can be distorted sometimes. My dataset/captionning was not great, so you often can see characters (or pieces of characters : hands or faces, especially the dad's) even when you don't prompt for them, even with the "solo" tag. The secondary characters aren't gread (the dad "nidad" is fine, but "alienman" can be weird, and "sacchan" is sometimes confused with the dad).
You need to specify a hair color, to determine which style you want to have. It's important, because Ninjin is seen with 2 hair colors in the anime. You don't need to specify the eye color (except if you want to change it, of course), because it adapts with the hair.
For the normal/orange style :
I recommend adding "blush stickers", as they are not added automatically (because the character doesn't have them every time in the anime). The quality of your result can improve if you add this tag, because the best images from the dataset have it.
The usual clothes are "blue overalls" and "green shirt", with "blue shoes" or "white socks" as well. Of course, you can chose any other outfit (like in the examples).
Normal/orange prompt =>
ninjin, 1girl, orange hair, blush stickers, blue overalls, green shirt,
For the night/brown style :
There is no specific rule. You may use "white nightie", but something like "pink pajamas" works well too, as it's shown in a (very) short scene from the anime.
Night/brown style =>
Ninjin, 1girl, brown hair, white nightie,
3 other characters are slightly trained, but the results are not great, as Ninjin is the main goal here.
Ninjin's dad, "nidad". You may add "glasses", "black suit" or "red tie"
"alienman"
"sacchan". You may add "blue jacket", but it's not really good so far. You should add "glasses" in negative prompt.
Note : every image in the examples is generated with this LoRA only (using AutismMix checkpoint) in Automatic1111, no ADetailer, nothing special. If you use the same parameters and seed than me, you should be able to replicate the images (tell me if you can't, please).
Description
First release version. The results are fine to me, but some troubles can occur with unexpected characters. I should clean the dataset and change my captionning for the next version.
Please feel fre to report any weird thing so I can try to improve it.
FAQ
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.









