Sign In

Flaring Coats and Bell-shaped+A-line Silhouettes

0

Not loaded

Download

1 variant available

fp16 SafeTensor

Flaring_Coats_1_epoch_8.safetensors

Half precision, best balance • 217.88 MB

Verified:

Type
LoRA
Stats

26

0

Generation

Not loaded

Generation License Fee

1 / 10 images

Reviews
Published

Oct 7, 2026

Base Model

Illustrious

Hash
AutoV2
80950E875B
default creator card background decoration
Followers - 654

654

Likes - 51

51

00089-3662804253.png

Hey you! Yeah, you. Care to make an entrance? Like a real entrance? I've got what you need...

!!!!!! EPIC COATS !!!!!!!

The answer to all your epic, cool entrance needs, and more!

But no, seriously. I wanted a lora to reliably do epically flaring coats and genuinely poofy dresses, and while it's not impossible to do those things with ai as-is(if you dedicate half your prompt for it), it's finnicky and hit-or-miss so I wanted a reliable shorthand so I would be able to have whatever else in the prompt. I'm not saying the lora is foolproof, no lora is, and obviously I'm still a bit of a newbie at this business, but I worked really hard this time to make something that I think works like maybe 85-95% of the time-ish. I (manually, as I do for all my loras... I honestly don't get the point of automatic tagging, how are you going to teach the model anything new if all your tags are concepts it already knows and recognises...but anyway) tagged 500+ images individually, from all sorts of angles and had to fiddle with the tags quite a bit. In the end I wound up training 4 loras, which ate up quite a bit of buzz(yes, I still do my lora training on civitai because despite having figured out how to do image generation and tagging locally I still can't for the life of me get kohya, onetrainer, ai toolkit or anything for lora training working locally XD... ) and now I'm honestly not even sure which to pick to upload, since multiple of them seem to have quite different strengths. I'll test them more thoroughly and upload more models on this page as I go. So, as usual, here's the ropes:

c0@7 = general trigger word

c0@7_vertical = does what it says on the tin. Flaring, but still mostly vertically falling coats/skirts/robes.

c0@7_horizontal = coats/skirts that are so strongly flaring they practically frame your character horizontally! Quite poofy. Which brings us to...

p∞f = for adding that extra ooomph. Can also be used combined with vertical for when you want a mostly gravity-affected coat/skirt that billows or balloons a bit at the bottom. Use with horizontal to, well, add oomph.

c0@7_stiff = for when you want it to look as if the skirt or coat is held up like that, and wouldn't lose its defined shape even when a character moves, think lolita petticoats and hoop skirts, as well as reinforced boning/pannelling.

c0@7_billowy, billowing = use to make it look as if it's only ballooning like that because of epic gusts blowing your character's silhouette into perfect shape. Usually these two go together, but you can also use them separately: use just billowing when you want to simulate the effect of an actually otherwise stiff coat or skirt in the wind, and use billowy by itself for when you want a coat/skirt that just looks like it was blown into the shape by wind but without necessarily the other symptoms of wind blowing touching the image.

c0@7_limp = bah. Gravity. Booooring. Use this when you wanna be a bore and get realism. Or, alternately, put it in your negative prompt for maximum poof!

c0@7_stationary = pretty self-explanatory. When it ain't moving. The coat/skirt I mean.

c0@7_curved = when there's this overall curve in the silhouette as it goes from top to bottom, great for more a-line silhouettes as well as for certain kinds of billowing, experiment and see for yourself, it's kinda difficult to describe.

c0@7_short = makes the length of the garment more on the short side. I find this one's a tad unreliable, so ideally I'd combine it with other stuff too to be sure. Not sure why. When it works it's great for lolita though. And for more like military coats.

c0@7_extra_long = well, whaddya think? Great for ballgowns and period-wear. Like in the re-enactment sense. What are you people thinking?!

pants_under, sl1t = use this if you want to ensure the outcome will be a coat that opens at the front rather than a skirt. What? What are you looking at me like that for?!

skirt_type, skirt = use to ensure outcome will be a skirt or dress.

3xag3r8 = use for maximum POOF!

t@1l_c0@7 = use for tail-coats and tuxedos.

m0t!onl3ss = for when you want no billowing whatsoever in your image. Hopefully. Most of the time.

bεl! = for particularily bell-shaped silhouettes.

sm∞7h = if you don't want any visible creases or folds in the fabric.

c0!1ar = just a cool kind of cape collar that frames the face that I found co-occurring in the dataset quite a bit. I've always loved these damn things so I gave them a custom tag and using it does make it somewhat more prone to producing them, though they weren't the point of the lora so it's not gonna be a guarantee. I might eventually make a lora that's all about these things.

Other potentially useful tags that aren't unique to the lora:

frontal, profile-frontal, profile, side view, from behind, profile from behind, low angle =defines the angle

detailed face, intricate background = if you wanna make sure there's enough detail in those. No extra loras needed.

plain background, concept art, multiple angles, = if you want solid color character sheet-type thingies. It can also do patterned background.

ornate, jeweled = very useful, for adding extra detail to the clothing. Each can be used independently.

stuff it also understands: hood, mantle, shawl, veil, headpiece, weapon, high heels, boots, bows, jabot, hat, kimono, sci-fi, off-shoulder, corset, leather, high specular, wide sleeves(put this one in negative if you want to get rid of sleeves interfering with the silhouette's definition), long hair, medium hair, updo, short hair, bald, dappled light, passers-by, etc.

Also worth noting if you use it to generate dresses: a weird thing I noticed is there's a tendency to make them hold the dress like in a curtsy, even though that wasn't much in the dataset. Might've also been something in my prompt or the checkpoint I tested with, but in any case I find it's mega-easy to fix by just telling it what they should do with their hands, crossed hands works if you're out of ideas. I also recommend using three legs or three limbs in the negative, as sometimes I noticed overfitting can cause it to insert an extra one.

And as always, tried and tested on both ladies and gents! :)