My first LoRA because I like Grea and didn’t see one for SDXL/Pony yet. <a target="_blank" rel="ugc" href="https://civarchive.com/models/282341?modelVersionId=629577">4th Tail Model v0.4.5</a> helps a lot for complex tail concepts like holding tail and more.<h3 id="tags-you-should-consider-adding-in-positivenegative:-cyqmpx4yc">Tags you should consider adding in positive/negative:</h3>See sidebar first for default outfit and bikini tags; Grea's Light outfit from GBF can be found <a target="_blank" rel="ugc" href="https://civarchive.com/models/758201">here as it's own LoRA</a> (Compatible with other characters) P: For v3 it's pretty much a requirement to add <code>dragon tail</code> in the prompt (see "About this version" for info)P: HAIRSTYLE: <code>short hair</code> or <code>long hair, hair over breasts</code> also <code>hair censor</code>helps. And: <code>ahoge</code> if you want it; looks cute :&gt;"Stabilizer" tags like <code>dragon wings, dragon horns, pointy ears, center frills</code> (and any others) not needed but use them if they are missing (model dependent); dragon-tags in negative might not work, untested; center frills if they disappear (usually when doing color change) N: <code>shoulder straps</code> when prompting high waist skirt (depends on model but they appear often from my testing because of the high-waist skirt)N: <code>breast pocket</code> to prevent said concept form appearing on the shirt/nightgownN: <code>lace-trimmed legwear</code> or (P)<code>plain thighhighs</code>if you encounter frilled/lace-trimmed thigh highs although from my testing they weren't nearly as bad as in v2 (i actually forgot to tag them in v3 so they got trained in, my bad)P/N: <code>high-waist</code> remove it or add it based on your preferences. Controls skirt somewhatP/N: <code>lace-trimmed bikini</code> (weak) for the blue-ish "fins"/lace-trim from the anime bikiniP: <code>anime screencap</code> could trigger the anime style depending on the model used. (untested on pony, but works very well with the <a target="_blank" rel="ugc" href="https://civarchive.com/models/826063">Illustrious</a> / <a target="_blank" rel="ugc" href="https://civarchive.com/models/826063">Tofu version</a> of this LoRA)Note: o-ring bikini might not work correctly in cowboy/full body shots because not enough data (merges with known but closeup/upper body should work <h3 id="flexibility-(colorsconceptsstyles):-nuonkteal">Flexibility (Colors/Concepts/Styles):</h3>Overall the LoRA should be pretty flexible. As in changing hair or eye color or style shouldn't be a problem although sometimes you need to add more weight or put the default color in negative when a color change isnt working. The high-waist skirt from default outfit seems to need extra work with color change. The wings seem to be trickier ("blue wings" for example made them completely blue but sometimes spilled over to the tail and "black wings" just kept the membranes at default color). Color change example (Black themed Grea): Here is an example of positive and negative tags i used for the black clothed/scaled grea (this was just a quick test to see if it's possible, i don't know if this is too many tags or not but it worked including the side ponytail hairstyle): P: <code>black horns, black wings, black dragon tail, black scales, black hair, dark purple eyes, black shirt, black high-waist, black pleated skirt, black plaid skirt</code> N: <code>red hair, purple hair, red plaid skirt, red high-waist, red pleated skirt, blue skirt</code>For more info and technical info, check "About this version".To support me without tipping Buzz, use the "Add post" button above the gallery (instead of "Post images" from the "Create" dropdown) if you want to publish images made with my LoRA. Each post earns me 50 Buzz. You can do this for other creators too.Side notes:<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/288584">AutismMix</a> sometimes gets the anime style quite accurate</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/448716?">BoleroMix</a> has great style and the author has a ton of clothing/concept lora worth checking out</li></ul>Update note 4th Oct.: There will be a 3.1 version at some point where i fix some of the tagging to cause less bleeding like o-ring bikini, but there aren't many images of the bikini so it might not even do anything. If I decide to work on a 3.1 then maybe I'll tag <code>tail under clothes/skirt</code> and see if it affects how other models treat Grea's tail and if it brings more control (see winter black capelet showcase image using t-ponynai3, tail is not under skirt and suggests back cutout/merge). If this test doesn't work out I'll just leave it as it is.

Grea (Multiple Outfits) | Manaria Friends / Granblue Fantasy / Shingeki no Bahamut | (Pony)

v3 and v2 should be better (and smaller in file size). If you are using this model aware of the newer versions please let me know why. Version info: First LoRA ever, cringe jpeg files from fancaps because they didnt have higher quality than 720p jpeg artifacted images (v2 has better images but they arent as clean and upscaled as well as bloated, v3 has the cleanest images i could get and debloated a lot) Problems:<ul><li>the long hair look Grea has in one episode is basically impossible from behind and front but is likelier to show if used with side profile</li><li>use <code>high-waist skirt</code> (or maybe<code>corset</code> too) in negative if you want other clothes, model seems to link that part to <code>grea</code> because im dum and dataset doesn't have high waist skirt captioned (fixed in v3 and probably v2, i dont remember lol)<ul><li>not sure if it's the lora or AutismMix/Pony models but it seems to have a bias towards military/marine-like clothes for some reason ( <code>badge, military</code> in negative seems to help)</li></ul></li></ul>Technical info:Trained on CivitAI LoRA trainer.Images were tagged using <a target="_blank" rel="ugc" href="https://huggingface.co/SmilingWolf/wd-convnext-tagger-v3">SmilingWolf/wd-convnext-tagger-v3</a>Showcase images were created in ComfyUI using a close to default workflow with a LoRA strength of usually 0.85 (sometimes 0.8 and rarely 1.0)

Files with SHA256: 2fcb8097d5779b0c477fa27a97071c0a7cf46f5f7ee84c47f97d2ba6ca0078b4