LORA by onpu segawa in magical doremi, learning through official and fan art.
Trigger: 'segawa onpu'.
It is also possible to change clothes by using the 'onpucos01' trigger for everyday clothes and the 'onpuwitch01' trigger for First season witch clothes.
Casual Trigger: 'onpucos01'.
1st season Witch Trigger: 'onpuwitch01'.
Weight: Between 0.6 / 0.8.
A setting of 0.6 will more easily reflect the characteristics of the checkpoint model, while a setting of 0.8 will result in a picture closer to the original work.
Example input:(casual)
(bloom lighting), (smooth lighting), (smooth coloring), (smooth shading), (light particles),(bestquality),(masterpiece), game cg, 4k,(5 fingers:1.5),(little girl), (ultimate cute girl), (segawa onpu), 1girl, (full body),purple hair, purple eyes, short hair, side ponytail, (onpucos01), (purple dress:1.5), (light green shirt), bike shorts, (yellow sneakers), (model pose), (smile), <lora:onpu_segawa_v1:0.8>
Example input:(witch)
(bloom lighting), (smooth lighting), (smooth coloring), (smooth shading), (light particles),(bestquality),(masterpiece), game cg, 4k,(5 fingers:1.5),(little girl), (ultimate cute girl), (segawa onpu), 1girl, (full body),purple hair, purple eyes, short hair, side ponytail, (onpuwitch01),(purple gloves:1.5), jewelry, purple hat, purple boots, earrings, compact (cosmetics), skirt, purple dress , (magical girl), (model pose), (smile), <lora:onpu_segawa_v1:0.8>
Example negative input:(common)
(flat shading:2), (flat coloring:2), (flat lighting:2), (multiple views:2), (worst quality:1.4), (low quality:1.4), (text:1.5) ,(trading card:1.5), (bad_prompt_version2:0.8), (bad-hands-5:0.8), watermark, signature, artist name, EasyNegative, (6 fingers:1.5), (7 fingers:1.5), (4 fingers:1.5),(logo:1.5)
Description
Illustrious 用として新たに学習。
Pony版に比べて原作に近いタッチで出力できるようになった。
New study for Illustrious.
Output with a touch closer to the original than the Pony version.
FAQ
Comments (6)
You know, if you combined all the characters into one LoRA, gave it 2 trigger words (one for the series and one for the character), this would would work well as a style LoRA.
Thanks for your suggestion. It would be possible if only style Lora were to be created.
In addition to that, it would be difficult to make Lora learn all six main characters accurately. There are also variations in witch costumes from series to series.
@waomodder It should be rather easy, as long as you make sure those tags don't get shuffled during training. It's how I did my own style & character packs. Then again, I never trained multiple canon outfits on top of the multiple characters, so maybe that would be too many layers?
If you aren't sure you want to put your own resources into it, DM me the datasets and I'll see what I can do on my end :)
I see. When I created my style Lora, I used the first tag as the trigger word and shuffled the rest of the words. But even so, Style Lora still worked just by invoking the trigger word.
If you are considering creating your own Style Lora, I can provide you with a dataset. In that case, I only have datasets for Hana and Onpu, and no other characters, is that OK?
@waomodder Oh, I thought you had more. Two characters may not be enough for a style LoRA to be consistent.
@AcanthAI
If you want to make a proper Style Lora, you will probably need four more costumes and character datasets.
If I were planning to make something again, I could try Style Lora next time.
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.