Lucina from Fire Emblem: Awakening, trained on animefull.
Around 0.8 weight works well, go lower if you want to change her outfit drastically. Trigger word is 'lucina fe'. Adding 'thigh boots', 'leggings', 'armor', 'shoulder armor', 'long sleeves', 'tiara', 'cape', and 'belt' can help with generating her default outfit. If you find it likes showing her thighs for no reason, try adding 'skirt' and 'zettai ryouiki' to the negative.
This was my first time working with such a large dataset for a character so I was kinda in uncharted territory here lol. Still, I'm pleased with how it turned out. It unfortunately cannot replicate her eye brand well, which is a little sad. Maybe I'll try training again at a higher resolution to see if that makes it better, but it'll also take me ages to do so :/. It also doesn't really do well with her mask, but you can get her short hair with '(short hair)'.
Description
FAQ
Comments (7)
I'm pretty motivated to do a bunch of Awakening LoRAs because I'm in love with the character design from that game, but LoRAs have already been made by others for some of the characters I'm planning on doing. Should I bother posting the duplicate LoRAs?
Go for it, dude. The old Lucina LoRA was pretty bad anyway. Maybe you could release a third Sumia LoRA along with me and TNP? ;)
@novowels Yeah I pretty much made this LoRA because I couldn't get anything remotely close to the example images using the old Lucina LoRA lmfao. As fun as it would be to have three Sumia LoRAs I don't think I'll do it because yours works great lol. I've actually been meaning to review it for a while but haven't got around to generating stuff to post lol. Definitely planning to do Cordelia, Chrom, Lon'qu, and probably the Robins for now. I'd also love to do one for Panne as well but she doesn't get enough art so it'll probably suck :(
@fireemblemfan By the way, I tried training a masked version of Lucina once, but the model absolutely refused to learn the mask. Ah well.
@novowels Huh, interesting. Maybe I'll give it a try to see if I have better luck but I doubt it. I'm actually really curious to see what it will do when generating the face if it's only trained using images with the mask on...
I'm on board, duplicate LoRas can be a net benefit if the original had far less fanart to work with, it's the logic I followed when doing my version of Goldmary's LoRa.
Edit: I know it's not the case with the other LoRas but still, if there's a room for improvement that's a good enough reason, more options is always good :)
@LordOtako Great points, I'll definitely post them then unless they turn out bad.
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.