Character Trigger Words:
nami \(one piece\), eyelashes, single sidelock, (narrow waist:0.85), long hairStyle Trigger Words (Select as Needed):
wano country arcegghead arcanime coloringcomictraditional mediaAttire Trigger Words:
blue bow, hair bow, back bow, blue kimono, yellow sash, short kimono, sleeveless, floral printstriped bikini, green bikini, lowleg pants, bikini top only, underboob, sideboob, midriff, denim, groinImage material sources: (~60%) screenshots from the One Piece animated series, (~10%) officially released images, (~10%) manga screenshots, (~10%) artwork by artist rider (orange couture), and (~10%) other fan arts. (This does not reflect the proportion of images in the training set.)
\
\
\
I must complain that the character designs in One Piece are extremely exaggerated, leading to many contents within images not being accurately recognized by automatic labeling models. If the training is focused solely on the art style, this wouldn't be much of an issue. However, if the goal is to accurately reconstruct the characters' attire and obtain clean backgrounds, manual intervention becomes necessary. Given that One Piece, as a series that has been running for several decades, features main characters with over a dozen different outfits, this task can be quite substantial.
Furthermore, the animation of One Piece from the Fish-Man Island arc to the Whole Cake Island arc suffers from numerous instances of character designs breaking down, which could potentially corrupt the consistency of the art style. Additionally, the quality of clothing-related tags within the One Piece category on Danbooru varies significantly, making it ineffective at distinguishing between two similar outfits.
Due to these reasons, the objective of this model is not to accurately reconstruct the attire..
Description
FAQ
Comments (9)
Super excited to try this out.
Can you upload the images with their original workflow?
Cause the images of this model are .webp and don't contain comfyui metadata. When trying to import I get (Unable to find workflow) message.
In my original workflow, there are many unapplied nodes, and the quality of the sample images is not the best because I used the PCM-16step model for acceleration (the PCM model has better versatility and compatibility than HyperSD, especially when using models that are quite different from the original SDXL). It's just a basic hires fix workflow.
@Recursion thanks for the response
yeah I managed to make it work, the results in this model are definitely cleaner in terms of skin, anatomy, and other body parts. It's even more accurate to my expression prompts now.
I have noticed however, the shape of the pupils is better IMO and more accurate to the anime source material in the previous nami model than in this model. Each have their own sets of pros and cons.
What checkpoint did you use to generate the samples?
Nice Lora. Works very good with the mosha style Lora.
Could you please update to include the elbaph outfit and hair? Would be awesome :)
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.