Original fictional adult character (18+). This character was created by Jimberly and is canonically over 18 years old. This LoRA represents a fictional adult, not a real person.
--
Alek Deng grew up in Ottawa, the daughter of a South Sudanese father and an African-Canadian mother. After finishing school and bouncing through a couple of jobs that never quite felt like hers, she moved east to Nova Scotia for a position at an animal shelter and discovered, somewhat unexpectedly, that she'd found her place.
Patient, warm, and armed with a very dry sense of humor, Alek genuinely loves animals while having absolutely no illusions about how ridiculous they can be. Her days involve cleaning kennels, arranging adoptions, taking profoundly uncooperative dogs for walks, and translating phrases like “high energy” and “very food motivated” into language prospective owners will actually understand.
She has a particular weakness for the elderly, awkward, three-legged, one-eyed, and generally unwanted cases. Alek insists she only fosters animals temporarily. Her steadily increasing number of pets suggests otherwise.
Away from the shelter, she's an enthusiastic recreational runner, haunts used bookstores, loves terrible horror movies, and enjoys cooking for friends. She'll happily spend an evening at home with a book and a dog asleep across her feet, although these days it's increasingly likely to be two dogs and a cat.
Sometimes two cats.
@doorknob22
Description
FAQ
Comments (18)
Help me, I fell in love with a LoRa! What can I do? @Jimberly this is such a well-done model. I only have one little point of critique: it seems that Alek really likes Jewelry, specifically earrings, so much that she wears them in any possible and impossible life situation and also gives them to every other character in the scene. ;-) Do you think it is possible to retrain the LoRa to make it a bit less jewelry-heavy?
Some people just get really attached to their accessories. What can I say?
I'm kidding, I'll see what I can do.
I've created a new Krea2, LTX2.5 and H3 LoRA overnight with the same dataset, just the images run through an edit workflow first to remove her accessories. So it should be virtually the same quality and output with the exception of the gold jewelery. Sadly, the AI overlord flagged her as a "real person", so we have to wait for the appeal to be approved, but you should be able to access her through .com for now, just not .red.
@Jimberly whoa! Thanks a lot, did not expect you to sit down right away. Much appreciated. I've tried "the new Alek" and while indeed the jewelry is gone, to me it seems like the images get a tiny bit more plasticky with the exact same settings and LoRas. Maybe its just me though or I am hallucinating. Or I am doing something wrong. Who knows. Anyway, I will figure it out. Thanks a lot agian!
@emkah Both are available now. Use whichever works best, I guess :D
Moss please?
I'm not sure what this means.
@Jimberly OP is talking about the new model Moss Image https://civitai.red/models/3005406/moss-image
@Tikiman Jiminy Christmas! There's another model to learn!
@Jimberly it never stops... Once you think you've mastered them all, somebody comes around with just another one...
Uploading now. Beta test, as this is my first MOSS LoRA. Someone tell me if it sucks. (But like.. be nice about it, I'm delicate.)
@Jimberly It works pretty good. Need strength at like 0.7 or it makes other people in the scene black as well with some face bleed. But at 0.7 it's good.
@phazei Ok, possibly I didn't have enough interaction shots then. I find if you don't put the character with different people in their originating dataset, the model starts to assume the character represents "all humans". There's other DOP tricks and stuff, but I haven't needed to lean on that too much.
@phazei I'll run it off again over night, check back late tomorrow.
@Jimberly you are hero. The way you react with lightning speed to requests is unmatched. Thank you for that. I hope you still have time to live a little. 😉
@Jimberly So, I've noticed one issue, it definitely decreases skin textures, things look less realistic with this. Maybe because the training images are like that? But it takes away from the realism that MOSS can create
@phazei that would be consistent with the same experience I have with the Krea 2 v2.0. Which is weird because @Jimberly told me he used the same image set for training as for the v1 version. I wonder if this is coincidental - it is another model after all - but my knowledge about training is very limited, so I cannot tell for sure.