Nico Robin from One Piece (Pre-Timeskip) in Crimson Style V3A complete remake of the character, using images of her appearances in J-Girl Impulse as the dataset.What do you think? Let me know in the comments!Nico Robin from One Piece (Pre-Timeskip) in Crimson Style V2 This version is the long-awaited update I mentioned — made after finding a model that fits Crimson’s style much better. As you can see, the result is way better now! It’s basically the same Nico Robin from Crimson’s works, but with a few extra outfits added.I worked on this version right away because I couldn’t leave the old one with the V1 issue. I hope you like this update! I’m not deleting V1 or its description, so anyone who wants can have a laugh and see how easy it is to make a few mistakes like I did in that first version.Nico Robin from One Piece (Pre-Timeskip) in Crimson Style V1 (obsolete) I need to tell a little sad story about this LoRA. To create it, I had to generate most of the images for the dataset, because even though the character appeared in Crimson’s works, she never appeared in any colored ones — and those early appearances were from Crimson’s old style. So, I thought, why not give her some of her One Piece outfits as well? (Yeah… a bit too ambitious on my part.)I tried to find a model that could handle Crimson’s style, but the first one had body proportion issues and didn’t adapt well to new styles. The second one worked better, but it had a different problem — it learned only the Enies Lobby outfit, and now it tries to turn every outfit into that one, as you can see in the images.I decided to release it anyway since it would’ve been a waste to throw away the work. When I find a better base model that can handle multiple outfits more easily, I’ll update this LoRA with a new dataset.For those who decide to use it despite this issue — thank you! P.S. If you’re good at prompting, please experiment with the outfits and share the results with me. It’ll help me understand how to improve this LoRA without having to rely on another one.I also want to publish LoRAs that didn’t turn out perfectly, because it’s part of the learning process — as the saying goes, you learn by making mistakes.

Files with SHA256: 1cef2ec31297aa68cbdb8d3266eb04990ced6c09789e3546162a896d2fddfbd5