(v3.5)New characters added:Aya: aya,lightblue hair,greyishblue eyes,low-tied long hair,blunt bangs,hime cut, very long hair, tsurimeShimamura Yu: shimamurayu, blue eyes, purple hair, short hair, hair ornament (cross shape),glasses,rimless eyewearSasa: sasa,one eye covered by hair,lightpurple hair,greygreen eyeYoung Kuesu: youngkuesu,lavander eyes,black hair,very short hair,crescent \(forehead birthmark)\,aged down,childKagamimori Kasuri: kagamimorikasuri,heterochromia,yellow and blue eyes,slit pupils,charcoalgreen hair,hair tubes (hair over shoulders),long hair(v3.0) Model name changed to just "Omamori Himari (TV Anime style)" and it is now a multi-character LoRA that can help generate images of the following characters from the show: Noihara Himari (Human, Catgirl and Ayakashi mode) Kuzaki Rinko Shizuku Jinguuji Kuesu Lizlet L. Chelsie Ageha Shama Kaya Kisaragi Sae (The teacher) Tamamo no Mae ("Tama") While it can get the appearance of the characters very accurately, the clothing and backgrounds are very inconsistent. Basic Appearance Prompts: Noihara Himari: noiharahimari,purple eyes,long hair,hair bow,ponytail,black hair,pink bow Noihara Himari (Catgirl): catmodehimari,purple eyes,long hair,hair bow,black hair,white cat ears, white cat tail Ayakashi Himari: ayakashimodehimari,white hair,black facepaint,red eyes,cat ears,slit eyes,black bodypaint,hair over one eye,very long hair,cat tail, evil grin, fangs Kuzaki Rinko: kuzakirinko,light brown hair,parted bangs,short hair,hair ribbon (red ribbon),orange eyes Shizuku: shizuku,green hair,hair intakes,red eyes,slit pupils Jinguuji Kuesu: jinguujikuesu,silver hair,white hair,lavander eyes,crescent \(forehead birthmark\),hair ribbon (black ribbon),gothic hairband, Lizlet L. Chelsie: lizletlchelsie,blonde hair,blue eyes,long twintails,maid headdress,hair ribbon,red ribbon Ageha Shama: agehashama,brown hair,dark green eyes,hair intakes,visible forehead Kaya: kaya,brown yellow eyes,dark hair,twintails \(two side up\),hair bobbles (\pink bobbles\) Kisaragi Sae (The teacher): kisaragisae,greeneyes,blonde green hair,short hair,glasses,rimless eyewear,lipstick,pink lips,makeup,hair bun,single hair bun,braid,mature female Tamamo no Mae ("Tama"): tamamo,yellow hair,drill hair,twin drills,aqua eyes,expresionless The new samples were generated using a custom model called <a target="_blank" rel="ugc" href="https://civarchive.com/models/118513/abased">abased</a> (v2.0) Updated LoRA with a bigger dataset and trained on NAI. Should be more consistent generating images of Noihara Himari, but it can also generate other characters from the show very closely. The trigger word for this version is promotionalart but it can work without it. (v1.0) I had first used SD 1.5 Pruned for training but It didn't replicate the art style that I wanted, then I decided to go with <a target="_blank" rel="ugc" href="https://civarchive.com/models/23900?modelVersionId=28562">Lykon's AnyLoRA - Checkpoint (Pruned)</a> for training which gave me just what I was looking for. I tested mostly on <a target="_blank" rel="ugc" href="https://civarchive.com/models/9409?modelVersionId=30163">AnythingV5</a> (works with other Anime style models) with 0.55 - 0.65 weight which gives me a very decent depiction of the character along with the art style, while also allowing some flexibility to change clothes poses and backgrounds, however since the training was done with only one image, the AI will attempt to generate something similar to the base image used for training, at anything above 0.8 weight, the AI will generate a similar pose to the base image more often while still allowing different clothing. No trigger word needed, but adding OmaHimaStyle seems to add a bit of extra detail (maybe is just extra noise, idk). This was mainly made to generate Noihara Himari images but surprisingly It can also be used as style LoRA when combined with other models, however this LoRA sems to start breaking with school uniforms when above 0.8 weight, likely because of the base image having a school uniform idk, I just wanted something quick to try out, I'm quite impatient. Enjoy (I guess?) but it can be fixed by adding more weight to specific tags. Basic prompt: Solo, 1girl, portrait, Black hair, purple eyes, very long hair, high ponytail, long ponytail, hair bow, pink bow, big hair, red shirt, sailor shirt, bangs, sidelocks, &lt;lora:OmaHimaStyle:1&gt;,

Omamori Himari (TV Anime style) -

Increased the size of the dataset with over 40 high quality images from promotional art for the show, this version should be much more consistent with the style, can work without a trigger word with a weight of 1.0 down to 0.6 without losing much of the art style. This LoRA's can generate Noihara Himari more accurately and also some of the other characters from the show.

Files with SHA256: 4b40ded37dbb9f41a286c23ec65e09ceeec294b593d2d11235df374a51939ab4