Inspired by great job of Master Corneo.
Not so perfect as his, but i tried.
So feel free to use and share your review in section down below.
Trained on NAI with clip skip 2, so it should work pretty fine with any anime-based model.
You also may use slight adjustments in prompt like adult, lady, big breasts, white corset, white hair, single braid, armored pauldrons to increase strength of embedding.
Also this negative prompt may be helpful twin braids, pointy ears, elf, loli
Description
More decent dataset, more training steps. So it's should provide more control.
FAQ
Comments (10)
Hi, I was wondering how I can get similar results. I'm using the same model with the same Hn and prompts but it doesn't come out, is there something I'm forgetting?
Hi. Sorry for bad english ahead, i'm not a native speaker. About your problem. Don't know for sure, but there can be some options like some of my hypernetworks have different name, so in my prompt it works and didn't in yours. So check name of your HN, embeddings and Loras. Also i mostly use clip skip 2 for generations, so you may check this as well. Also my GTX 1660 Super isn't super fast despite it's name, and because of that i sometimes use this pipeline: generate batch of some pics (about 4-6) without hires.fix, then choose best, use same seed with high.res fix about 1.5, and then send it to img2img. And then in img2img with denoise about 0.3-0.4 and slightly higher resolution i make some adjustments, like changing HN, changing clip skip and other stuff. It's all costs me about 5-10 minutes for one decent pic. And this also the reason why sometimes my generation data didn't match the data from info tab on civitai. But in this TI page i mostly tried to generate directly through img2img and hires, without any fancy ways, so my generation data and prompt info actually match.
@sadxzero Setting clip skip to 2 worked for me! Thank you, with Clip skip 1 she ends up looking like a generic anime girl.
@petdragon very glad, that you figured it out, enjoy)
Privet! ;) Hey, since you started making your own textual inversions, i have a request. Corneo's list are pretty big so waiting for it gonna be a long a time. The request is.. Raiden from Metal Gear Rising. It's a blonde cyborg guy. That'd be awesome if you accept it :)
Привет) Oh, I can't promise anything decent, so don't expect too much of me. I still trying to learn all this stuff. But I will try to make Raiden on this week.
Would love to use this, but unfortunately only getting pretty flat and old ladies. Not sure what I'm doing wrong. I'm using the trigger word, the extra given words and negative prompt using the v1.5 pruned ema only model. Am I using the wrong model?
Very nice! Need to try this out and review when I get the chance and remember!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.