We will comply with the ANYCOLOR secondary creation guidelines.
https://www.anycolor.co.jp/guidelines/

It seems that no particular intention is indicated. I pledge to create and publish works that respect the spirit of ANYCOLOR content.

特に個別の意向が示されていないと思われます。ANYCOLORコンテンツの精神を尊重した作品を制作、公表していく事を誓います。

We will comply with the ANYCOLOR secondary creation guidelines.
https://www.anycolor.co.jp/guidelines/

It seems that no particular intention is indicated. I pledge to create and publish works that respect the spirit of ANYCOLOR content.

It seems that no particular intention is indicated. I pledge to create and publish works that respect the spirit of ANYCOLOR content.

BigrBear | Stylized「LoCon」

Training models take alot of time and effort! Support me by leaving Feed backs and Reviews!☕

This is Lycoris Network as such you will need an extension to utilize them.

(I recommend using it natively instead of loading the network via KohyaSS Extension as the Kohya extension likes tobreakat times.)

My models are made to be used at [ Weight 1 ] Unless it is being combined with other Networks.

There are 2 Versions to this model.

「Normal」and「Hard」

「Normal」is very flexible and you're able to prompt just about anything with it.

「Hard」is Fixated to the dataset and will create backgrounds very similar to the dataset with the artists stone like texture on every background image. Its not recommended to use but its there for the sake of being there.

TOS! Terms of Service!

• You are not allowed to sell generated images

• You are not allowed to host this model on generation services

Exceptions : Unless the style of this model is merged enough with different styles making it not easily distinguishable then the TOS does not apply to such derivative models.

Here's a google co to run this LoConalong with other models If you don't meet the requirements to run Stable Diffusion Locally.

We will comply with the VERSE secondary creation guidelines.
https://verse-n.com/guideline/
Please read the secondary creation guidelines.
二次創作ガイドラインをご一読お願いいたします。

We will comply with the VERSE secondary creation guidelines.
https://verse-n.com/guideline/
Please read the secondary creation guidelines.
Please read the secondary creation guidelines.

just for fun.I capture from low res video.just feel this man is very funny.so don't care.

John Doe -A Style Lora

This is more or less an experiment in creating Lora/Locon, and not meant to be a replacement or a strict recreation of John Doe's work. Please support the original artist!
https://www.patreon.com/JDArt

This is a style Lora (a Locon, specifically. So you'll need the extension to properly use it) meant to emulate the style of John Doe (dimensional forms, dramatic lighting, "realistic-er" facial features). That's pretty much it.

Seems to work best for me at around 0.9 strength, but feel free to play around with it. Lower settings will flatten things out and bring out more "anime" features (though certain tags like "smirk" tend to have a strong effect of bringing those out as well).
It was trained on explicit images as well, so it should be pretty fine generating actual intercourse, and most subjects therein. Though you might have the occasional floating dick here and there.

While the Lora also does well enough as you can expect at 512x512, I recommend using a slightly higher resolution to bring out facial features a bit more. Either by using a higher base res, or just turning on hires fix.

Since this was kind of a "lazy" Lora that I didn't meticulously curate, things might get a little off from time to time. Your mileage may vary, but the age-old adage reigns true, the better your prompts, the better your result. Still, it's absolutely possible to get decent results using basic prompting.

Feel free to leave any comments you've got as well as any images you have made!

The Lora is made using my usual workflow of training 2 separate Lora. One on full bodies, and another with head crops, and merging them together.

Nobunaga's Ambition is a series of turn-based grand strategy role-playing simulation video games. The original game was one of the first in its genre, being released in March 1983 by the Japanese video game developer Koei. Nobunaga's Ambition takes place during the Sengoku period of feudal Japan. The player is tasked with achieving the ultimate goal of warlord Oda Nobunaga: the conquest and unification of Japan.bla bla bla,hey man just take it!

original description

A Character LoCon for Reki Kyan (喜屋武暦) 【sk8 the infinity】

Weight:0.4～0.6

Hires. fix

trigger words: kyan reki\(sk8 the infinity\)

※以下日本語で失礼します。

運要素が絡みますがぽい衣装を出現させることは確認しています。

【注意】スケボが題材なのにスケボに乗った画像生成はレアです。スケボに対して暦くんがいい位置に来ない、スケボが2枚出現する、スケボが異様に小さい、ローラーが６つ生成、靴からローラーが生えてくる、スケボが曲がっているなど…t2iで200生成近くテストしましたが指ガチャも合わさり根気がいります。体感512*768よりも上のサイズ640*960などだとまともになりやすい気がします（高くしすぎるとSD特有の奇形体型になる）。

どうしてもスケボ暦を作らせたい場合は大人しくcontrolnetかi2iで修正を加えつつが大正義かもです…。また学習の過程でグラフィティアートとスケボ柄をユーザーIDやサインとご認識している感じがありますので気になる方はネガティブにuseridなどお書き下さい

共通呪文 : kyan reki\(sk8 the infinity\), red hair, short hair, orange eyes, headband

※orange eyesはyellow eyesとお好みで使い分けて下さい。第7話のあのシーンを再現したい場合は下記にも書きますがheadbandを消しておいて下さい

デフォ衣装

出現率アップ呪文:black gloves, hoodie, wristband, yellow hoodie, black jacket, black pants, school uniform

※靴が欲しい場合はshoes, sneakers, 追記。基本青色が出現すると思いますが出ない場合はblueを足して下さい。

私服

出現率アップ呪文: 〇〇 hoodie,

※基本パーカー属性の彼なので〇〇に色指定。

7話のアレ（髪下ろし）

出現率アップ呪文: hair down, hair between eyes

※プロンプトのheadbandを消しておいて下さい。ネガティブにheadbandを書かなくても成功率100％だったので多分大丈夫かなと思います。

以上です！！

最後に、暦くん以外のSK∞ エスケーエイト他のキャラLORAもしくはLOCONを作るつもりはないです。

A Character LoCon for Reki Kyan [sk8 the infinity]

Weight:0.4～0.6

Hires. fix

trigger words: kyan reki\(sk8 the infinity\)

※The following Japanese words are not rude.

Luck element が结みますがPoiClothing を appear させることはconfirmationしています.

【Attention】The theme of Skateboarding, but the portrait is generated is rare. 6 are generated, rollers grow from boots, skateboards are bent, etc... I tested t2i de200 Generating near, but it takes patience because it also includes gacha. I feel that 960 etc. is easy to get decent (if it is too high, it will be a strange shape of SD unique).

I really want to make skateboards.I think it might be a big justice to add correction with controlnet or i2i... Also, in the academic process, I feel like I know graffiti art and skateboards as user IDs and signatures. If you are interested, please write your userid negatively.

Common 呪文: kyan reki\(sk8 the infinity\), red hair, short hair, orange eyes, headband

*Please separate the orange eyes from the yellow eyes if you like. If you want to reappearance that scene from Episode 7, I will write it below, but please remove the headband.

default clothes

Appearance rate up sentence: 〇〇hoodie,

*Since it is a basic hoodie attribute, color designation is given to 〇〇.

That in Chapter 7 (Haircut)

Appearance rate increase sentence: hair down, hair between eyes

* Please turn off the headband of the prompt. The success rate was 100% even without writing negative headbands, so I think it's probably a great teacher.

The above desu! !

Finally, I have no intention of making Koyomi-kun's SK∞ SK8 character LORA or LOCON.

We are all love ALIEN right?

Add at 1.0 weight, prompt sarashi or fundoshi. Base anime models couldn't do them well enough so I trained this.

Sarashi are the chest wrap, fundoshi are the underwear.

True Pilyeon

Model Made on the Art Generated from the previous model.

This model shouldn't violate any kind of copyright or delusional/Ethical/Self-made/Unethical laws.
Please, I'll appreciate your support on my Patreonandkofi.

Here is a Form you can request me Lora there (for Free too)

https://forms.gle/h9iuLCCnaEs6W9ZB8

This is a model of Jesus Christ from the show Black Jesus.

Use jesus-christ as your tag and it should work. Seems fine at a power of 1 but I'll do some testing. This one is fresh out just for Easter.

We will comply with the ANYCOLOR secondary creation guidelines.
https://www.anycolor.co.jp/guidelines/

It seems that no particular intention is indicated. I pledge to create and publish works that respect the spirit of ANYCOLOR content.

特に個別の意向が示されていないと思われます。ANYCOLORコンテンツの精神を尊重した作品を制作、公表していく事を誓います。

We will comply with the ANYCOLOR secondary creation guidelines.
https://www.anycolor.co.jp/guidelines/

It seems that no particular intention is indicated. I pledge to create and publish works that respect the spirit of ANYCOLOR content.

It seems that no particular intention is indicated. I pledge to create and publish works that respect the spirit of ANYCOLOR content.

I would suggest using "multicolored hair, streaked hair, abstract, colorful" for the full effect.

「This figure is... ah, that's what it is. If this happens, let's stop hating the demon's blood flowing through our bodies from this time today. Tomoe fights with all the power she has」

• CFG Scale: 5.5 - 7

• Weight: 0.8 - 1

• Recommended prompts: To Be filled. (See sample images for now)

✨✨Want your personalized LoRA or illustrations? Click here✨✨

Want your own LoRA and illustrations? click here

🎨 Pixiv https://www.pixiv.net/en/users/22936156

🐦 Twitter https://twitter.com/tsumasaky

📷 Instagram http://instagram.com/tsumasaky

Note #1:Please read the "how to use" section for this Locon. While it does a lot of the heavy lifting to get a good picture, it definitely still needs a little help from generic tags.

Note #2:This is a Locon. Install the extension to get it to work. I've found them to do better when working on ideas that I want easily added to one of my LORA characters. Also, they're WAY less damaging to images, even at high weights.

After many, many version of this, I think it's in an okay enough state to release. So, here it is, my brainwashing Locon. It's still not perfect and could use a v2 (and v3, and v4, etc.) but I've made 9 versions so far and I just want to release something lol. It was trained on animefull_final with 101 images with the usual self-pruning and tagging and a wide range of different images. I like keeping my prompts simple for the first few pictures so that you can see it without a whole bunch of mumbo jumbo... I would say. This one is a teeny different. Make your pictures better than mine though!

What does it do you ask? A Locon that will create a "brainwashed" waifu who has the general blank stare and lustful. Sex optionally can be included.

How to use it: I've found that with these Locon's I've been submitting, weight's can be wide and varied. This one specifically can range from 0.6 to 1.4 in terms of how much you want it to take over the image.

Ok. Now trigger words. First, baked into the Locon we have:

BrnWsh -Main word if you just want to get the thing to be recognized well. Use it hwen you don't care which of the next option you get AND when you want to force the look even more-so.

NFucked -To lean the Locon to create images where the character is not having sex.

BotFucked (recommended 512x768 width x height) -To lean the Locon to create images where the character is being fucked while on top of someone.

SFucked (recommended 768x512 width x height) -To lean the Locon to create images where the character is being fucked from the side.

ADDITIONALLY USE THESE:

1boy, hetero, penis (or dildo), sex, vaginal (or anal or oral) - I like pairing my images with these to increase the chance of success of the image.

OPTIONALLY USE THESE:

glowing eyes, rolling eyes, swinging pendulum, fucked silly, clothes lift - Or anything else you want to be emphasized. It should adapt to what you add.

IMPORTANTLY, EVEN WITH ALL THIS SOME PICTURES WILL JUST BE BAD :(.

Extra, Extra!: Should be well trained and usable in most models. Images shown for example.

Picture Descriptions:

1-3: AnythingV3 using just this Lora with simpler prompts

4: AnythingV3 combo'd with my in progress V2Corruption Locon.

5-6: AnythingV3 combo'd with my released Seraphina (Disgaea 5) Lora.

7: AnythingV3 combo'd with my release Ghislaine (Mushoku Tensei) Lora.

8: AnythingV3 combo'd with my in progress v2Ameri (Iruma Kun) Lora.

8: AnythingV3 combo'd with my in progress "Symbioted Taking Over" Locon (basically a symbiote merging with the user).

9: AbyssOrange with just this Lora.

10: PerfectWorld using just this Lora.

What's next for this? Find ways to make this better. Also, bring back the BoobFucked and SPendulum tags that I had to cut since they weren't helping much.

What's next that isn't this?

More character Loras. (Ivyfrom Soul Calibur 5, Heinefrom Combatants Will Be Dispatched, I conquer Lilias from E7, Rozalin from Disgaea 2, Dry from Fire Emblem, Catherine from Catherine Full Body, Mikufrom Idaten Deities, Kuribayashi Shino from GATE, Sohma too from Fruits Basket, Kasumi from P5, Ann from P5, Ikumi Mito from Food Wars, Alice I confess from Food Wars, Chika Fujiwara from Kaguya-sama, Lulu from FF10, Momo from My Hero,Vivi from Vivy, Dominique de Sade from Vanitas, and Of the from One Punch Man.) And maybe William of Astrea, my favorite Re:Zero husbando or aPrinny from Disgaea.

V2's of some of the previous Loras. Specifically make them more diversely acceptable with other models and LORA's.

I also have some more quite NSFW Locon's. I decided to list them as well, so if people have preferences on which to fix up first let me know. Specifically, CorruptV2, Symbiote Taking Over (imagine a venom-like symbiote taking over your waifu), being stepped on, Robotification, Petrified (I'm unsure if I should bother with this one anymore), Infection, Dominatrix (waifu becoming a dominatrix), Brainwashing Machine, Tentacles, Pillory, Masked Evil(A mask that possess' the waifu), Latex Corruption, Hourglass Figure, Combat Drone (hard to describe but basically making them a mindless sci-fi combat warrior), JAV Supervillain, Jiangshi,andHolding a Leash.

And, uh, I want to make a few "normal" Locon's so I have an "knight-ify" your character one coming up. And a forest fairyone as well. And a officer one. And a barbarian one.

Or just give me some random suggestion or priority opinions. That works too.

A lycoris model of bed invitation pose.

Trained on 55 anime pictures on NAI model.

Trigger: bed_invitation.

Samples are generated on Anything4.5 with LoRA Block Weight.

Introduction

A Lycoris model trained with ~20 images of Medusa from Battle Through the Heavens. The trigger tag is ‘c-medusa’.The 784mb VAEs (NovelAI, Anythingv3, Orangemix, Counterfeit) are recommended. 0.6 ~ 0.8 weights are good. The preview images are generated using 1.6 version (original)YesMix. This model is a bit overfitted due to few training images, which will be fixed when dataset is updated.

Cautions (Must Read)

• Some preview images are generated using hires fix. Thus, if you wanna generate the preview images, pls download the original preview images and analyze them using 'PNG INFO' in webui to get all generating information.

• The maximum of generating resolution should not exceed 1024 theoretically, otherwise extra/deformed head/limbs will be waiting for you :).Try to use 'hires fix' to prevent this problem and help to generate higher-resolution images.

• For 784mb VAEs (NovelAI, Anythingv3, Orangemix, Counterfeit) users, make sure add '--no-half-vae' in command line of webui to prevent generating black images. See voldy's log for the details of edit method. For high-performance GPU, make sure 'medvram' and 'lowvram' mode are disabled.

Mellow swimsuit worn by Ayane from the Dead or Alive franchise.

• Trained with Anime (full-final pruned) model.

• Recommended using with inpainting by masking the desired areas for the outfit to appear.

• Recommended making a image of a character wearing a green two-piece swimsuit (with short included) before applying this LoRA with inpainting.

It's a LoRA that makes you look like an anime-style Asian for live-action.

It may be better to use it for adjustment in combination with other face LoRA.

No stratified LoRA was applied.

~~~Translated by DeepL~~~

This LoRA can resemble an Asian person in a anime style.

It may be better used in conjunction with other facial LoRAs for adjustment.

No lora block weight are used.

~~~~~~~~~~~~~~~~~~~

weights: 0.6~0.8.

trigger words: none. (1girl)

We will comply with the ANYCOLOR secondary creation guidelines.
https://www.anycolor.co.jp/guidelines/

It seems that no particular intention is indicated. I pledge to create and publish works that respect the spirit of ANYCOLOR content.

特に個別の意向が示されていないと思われます。ANYCOLORコンテンツの精神を尊重した作品を制作、公表していく事を誓います。

We will comply with the ANYCOLOR secondary creation guidelines.
https://www.anycolor.co.jp/guidelines/

It seems that no particular intention is indicated. I pledge to create and publish works that respect the spirit of ANYCOLOR content.

It seems that no particular intention is indicated. I pledge to create and publish works that respect the spirit of ANYCOLOR content.

This LORA is Kashii Airi from RO-KYU-BU!

initial tags:

"kashii airi, short hair, brown hair", (large breasts)

main costume:

"gym uniform, bike shorts"

"basketball uniform, shorts"

"hoodie, shirt, shorts"

You may want to use 7thLayerAbyss or AOM3A.

show you guts cool say what saikou daze.

Making loras takes up a part of my free time, any help is welcomed: https://ko-fi.com/truefalse

Sometimes in rear view it wants to make a different vintage car, don't know why
Also it's a shame i didn't manage to get the ai to generate the big beautiful multiple exhausts this car has...
image grid: https://files.catbox.moe/tks9q0.png

I check Twitter every day, so I will respond when I receive a DM.
ツイッターは毎日チェックしているのでDMが来たら対応致します。

I check Twitter every day, so I will respond when I receive a DM.
I check Twitter every day, so I will respond when I receive a DM.

2023/4/9 update

Update dataset. Upload lycoris version. v1.0 version is conventional LoRA model. The preview images are generated using 1.6 version (original)YesMix.

Introduction

A conventional LORA model of musashi (azur lane) trained using 160 highres images, including nsfw. The NAI's VAE is recommended. 0.6~0.8 weights are good.

Cautions (Must Read)

• Some preview images are generated using hires fix. Thus, if you wanna generate the preview images, pls download the original preview images and analyze them using 'PNG INFO' in webui to get all generating information.

• The maximum of generating resolution should not exceed 1024 theoretically, otherwise extra/deformed head/limbs will be waiting for you :).Try to use 'hires fix' to prevent this problem and help to generate higher-resolution images.

• For 784mb VAEs (NovelAI, Anythingv3, Orangemix, Counterfeit) users, make sure add '--no-half-vae' in command line of webui to prevent generating black images. See voldy's log for the details of edit method. For high-performance GPU, make sure 'medvram' and 'lowvram' mode are disabled.

We will comply with the with the secondary creation guidelines.
https://vspo.jp/guide/
Please read the secondary creation guidelines.

Since it is a lokr model, please use the latest version of webui and the extended function locon.

We will comply with the with the secondary creation guidelines.
https://vspo.jp/guide/
Please read the secondary creation guidelines.

Since it is a lokr model, please use the latest version of webui and the extended function locon.

Retrained with LoCon using Hollowstrawberry Colab

• Smaller size 28.9mb

• Better costume

original description

<lora:human_anatomy_v1-9:0.7>,human_anatomy_style

the character should expose some skin without clothes.

,human_anatomy_style

the character should expose some skin without clothes.

A Character LoCon for Will.Sprout [Helios Rising Heroes]

Weight:0.4～0.6

Hires. fix

trigger words: will sprout\(helios rising heroes\)

*Please excuse me for the following in Japanese.

運要素が絡みますがPoi衣装を出現させることは確認しています。

Common Spells: will sprout\(helios rising heroes\), blonde hair, short hair, brown eyes

uniform

Appearance rate spells:collared shirt, belt, black shirt, black pants, belt buckle, yellow necktie, police uniform, helios uniform,

hero suit

Chance Spells: black gloves, gloves, black shirt, yellow gloves, goggles, goggles on head, two-tone gloves, yellow jumpsuit, headset, yellow coverall, the kindhearted green thumb

*The kindhearted green thumb is a charm spell

Swimsuit (Midsummer Survival Island)

Chance Spells: shorts, sleeveless, hood, fingerless gloves, goggles, goggles on head, swim trunks, sleeveless jacket, yellow jacket, hoodie, diving_mask, diving mask on head, Survival Island

* Survival Island is a talisman. She has a habit of opening her jacket. Please play gacha or control with spells. Midriff peeks like in the original... (difficult to control)

Kaho from Blue Archive

• Trained on YesMix 1.6

• Weights between 0.5-0.8 seem good. Use higher if you want to use with original clothes.

• Careful with colours, likes to bleed into clothes.

A model trained on _himehajime and their style of artwork.

魚介 @_himehajime). https://www.pixiv.net/en/users/165192

Note:

This is a LYCORIS/LoCon model, and you need to install a1111-sd-webui-locon extension to use this model.

This is a LoRa with a vector graphic art style, which tries to showcase simplicity and flatness in colors as much as possible.

There are two versions available here, namely the trim version and the original version. For specific differences, please refer to the version introduction. Generally, downloading the trim version is sufficient.

Trigger word:flat color

Suggest weight:1

I really like the color scheme of this LoRa, and by the way, you could add chibi tag to make the character generated even cuter and more distinctive.

Based onnovel-anime-full model.

Since SuperMerger currently does not support merging LYCORIS, I may need to create a LoRa version if necessary.(But generally, I wouldn't do that.)

Neo Universe from Uma Musume

This is a LoHa LoRA! You need the extension!
Why? LoHa have virtually no effect in output art style and they do a lot better with multi-concepts.

Tired of cute anime faces? Me too. Lets fix that!

Trigger wordsfor V1.20:

positive: (mature female)

negative (for less muscles in the body, if you prefer that): muscular, toned

You can also try "lips apart", "small breasts", "thick thighs" and "wide hips" also helps out with more mature look!

Follow the steps in the Model Version notes to reproduce my images.

Sample images generated with Hassaku.

All negative embeds should be easy to find. and Prompts should be included in the sample images! Click the little icon in the bottom right to see it.

It was trained on 122 512x512 images borrowed from Unstable Diffusion discord, 20 epochsusing Animefull-final-prunedfor the Training Model.

Danbooru tags generated by Waifu Diffusion 1.4 Tagger.

I made it for another purpose, it's pretty easy to accidentally call a bear
Steps: 50, Sampler: DPM++ 2M Karras, CFG scale: 4
don't call dog 😂

I made it for another purpose, it's pretty easy to accidentally call a bear
Steps: 50, Sampler: DPM++ 2M Karras, CFG scale: 4
don't call dog 😂

WARNING: This is not meant to be used to deepfake this brilliant man. IF anything this file is a mere tribute. (Queue Jack black and Arin Hanson singing) - Yes, i'm sure you could produce your own only fans, but let's be honest - Why would you?

This is a FAN created piece, and yes sure Mark I've literally taken screenshots off your videos - but y'know I don't intend on doing anything dirty with this (and I would sincerely hope nobody else does LOL).

Best tips: Use an anime model at about .7 strength! Comes up with GREAT AI fan art ;)

NO THIS IS NOT MEANT TO REPLACE ART OR FAN ART :3 I just ... I wanted to see if it could do it LOL.

THIS DEPICTS MARK FISCHBACH OR MARKIPLIER.

What you do with the pictures is up to you, I am not advocating or anything to do with the final product of what you make.

However, PLEASE LET SOMEONE TEACH MARKIPLIER WHEN HE HAS TIME TO DO STABLE DIFFUSION ON SCREEN - I really want him to use this. (Or Jacksepticeye, and then have him like laugh maybe... i dunno just don't sue me pls XD)

Urethral Blowjob!! Peehole Blowjob!!👍

LoHA model for TRS sister. It is seems like the background is overbaked so you may add more prompt to background, idk.

We will comply with the with the secondary creation guidelines.
https://vspo.jp/guide/
Please read the secondary creation guidelines.

We will comply with the with the secondary creation guidelines.
https://vspo.jp/guide/
Please read the secondary creation guidelines.

Anzu Mazaki/ Téa Gardner from Yu-Gi-Oh! Duel Monsters

• Trained with Anime (final-full pruned) model.

• Comes in 3 versions; 6 epochs, 8 epochs, and 10 epochs.

• You can use the following activation tags;anzu_yugioh:for general info,anzu_classic:to get her classic hairstyle,anzu_dsod:to get her Dark Side of Dimensions hairstyle (make sure to typeshort ponytail, high ponytailto get her ponytail in the movie.)

Christmas Kiss/アマガミ/amagami style

No keyword set for role

You have not set a trigger word for your character

No trigger words are set for the characters.

The following character names can be used to generate the corresponding characters, which are not guaranteed to have the same effect as the character lora

The following keywords can generate corresponding characters, but cannot guarantee the same effect as the character Lora

It can be generated using the following character names, which are not guaranteed to have the same effect as the character Lora.

ayatsuji_tsukasa, morishima_haruka, tanamachi_kaoru, tachibana_miya, nanasaki_ai, sakurai_rihoko, kamizaki_risa, nakata_sae, tsukahara_hibiki, takahashi_maya

The example image is generated using t2i -> i2i -> Tiled Diffusion+Tiled VAE, and the prompt other than the image quality is deleted in the last step

Sample diagrams are generated in the order of t2i -> i2i -> Tiled Diffusion + Tiled VAE, some prompts are deleted when Tiled Diffusion + Tiled VAE

Samples are generated by t2i -> i2i -> Tiled Diffusion+Tiled VAE. Some prompts were deleted by Tiled Diffusion+Tiled VAE process, so the prompt of the samples is incomplete.

Samples full pormpt:

1. (masterpiece, best quality, extreme detailed beautiful eyes), best illumination, perfect lighting,look at viewer,(1girl, solo, close-up:1.3), dynamic angle,bedroom, (smile:1.3), sitting on bed,
   (kibito high school uniform:1.3),black jacket, pleated_skirt,school_uniform, blazer, large breasts, collarbone,thighhighs, no shoes, white hair, ponytail, long hair, yellow eyes,
   

2. (masterpiece, best quality, extreme detailed beautiful eyes), best illumination, perfect lighting,look at viewer,(1girl, solo, close-up:1.3), dynamic angle,bedroom, (smile:1.3), sitting on bed,
   (kibito high school uniform:1.3),black jacket, pleated_skirt,school_uniform, blazer, large breasts, collarbone,(nanasaki ai:1.3)
   

3. (masterpiece, best quality, extreme detailed beautiful eyes), best illumination, perfect lighting,look at viewer,(1girl, solo, close-up:1.3), dynamic angle,bedroom, (smile:1.3), sitting on bed,
   (kibito high school uniform:1.3),black jacket, pleated_skirt,school_uniform, blazer, large breasts, collarbone,
   , ,arianeglenysmaple,white hair, pointy ears, ponytail, hair ornament, yellow eyes,

4. (masterpiece, best quality, extreme detailed beautiful eyes), best illumination, perfect lighting,look at viewer,(1girl, solo, close-up:1.3), dynamic angle,bedroom, (smile:1.3), sitting on bed,
   (kibito high school uniform:1.3),black jacket, pleated_skirt,school_uniform, blazer, large breasts, collarbone,
   , ,sashanecron,twintails, blonde hair, blue eyes,

5. (masterpiece, best quality, extreme detailed beautiful eyes), best illumination, perfect lighting,look at viewer,(1girl, solo, close-up:1.3), dynamic angle,bedroom, (smile:1.3), sitting on bed,
   (kibito high school uniform:1.3),black jacket, pleated_skirt,school_uniform, blazer, large breasts, collarbone,
   , (white hair:1.4), sidelocks, ponytail, open jacket, open clothes,

Hikage from the Senran Kagura franchise.

• Trained with Anime (full-final pruned) model

• Comes in 2 versions; 3 epochs and 4 epochs, use which ever fits you the most. I also advise you to make combinations of them to get interesting results.

• Try them with any LoRA block weight.

We will comply with the with the secondary creation guidelines.
https://vspo.jp/guide/
Please read the secondary creation guidelines.

We will comply with the with the secondary creation guidelines.
https://vspo.jp/guide/
Please read the secondary creation guidelines.

If you'd like to support me and do more:
https://ko-fi.com/edgonkofi - commissions open!
If you're looking for a >>LoRA Making Tutorial<<

Hard Swimwear

This one is for those who don't know if they want to have a swim or to have sex.

Why not both? Get this LoCon now!

A LoHa trained on a measly 10 images using my method. 1 batch size.

This was me experimenting with a plain 1boy style dataset to get an idea of how fast the AI should learn those without overfitting. Uses my own methods. Dataset was provided by somebody else, I hadn't done anything to it except remove 16:9 aspect ratio photos.

27 image dataset.

Just a little extended lora trained with a small dataset of ~130 images.

Dataset is 100% SFW, so it will not help you produce NSFW imagery, but it shouldn't tamper with that ability either, as it's trained with enough classification images and shouldn't bleed in any meaningful way to all your lewd tags.

It's triggered with javi token. Without it effect should be negligible, or not appear at all.

Dataset was gathered from MJ with specific style prompt, though, v5 models are quite rigid(especialli niji), so, im not sure it made something unique enough. But one thing is for sure - it's not relevant to any artist.

Basically, whole thing of this one is i needed to waste my MJ subscription hours, okay? If you would enjoy using it - good xD

If you create any good images using this LyCORIS, please consider sharing them by leaving a review. Thank you!

• This LyCORIS can help generate specific stewardess uniforms Including red and dark blue uniforms.

• This is a Lycoris, please make sure you have installed the related extension.

• Stewardess, Red uniform, and Dark blue uniform to trigger.

• This model cannot accurately reproduce the appearance of Russian people; it mainly provides the styling of the uniforms.

• If you want to remove the hat, put "garrison cap" and "hat" in the negative prompts.

• The scarf is considered part of the uniform and is difficult to remove separately.

Introduction

A Lycoris model trained with ~260 images by General. The trigger tag is ‘jeneral’.The 784mb VAEs (NovelAI, Anythingv3, Orangemix, Counterfeit) are recommended. 0.6 ~ 0.8 weights are good. The preview images are generated using 1.6 version (original)YesMix.

Cautions (Must Read)

• Some preview images are generated using hires fix. Thus, if you wanna generate the preview images, pls download the original preview images and analyze them using 'PNG INFO' in webui to get all generating information.

• The maximum of generating resolution should not exceed 1024 theoretically, otherwise extra/deformed head/limbs will be waiting for you :).Try to use 'hires fix' to prevent this problem and help to generate higher-resolution images.

• For 784mb VAEs (NovelAI, Anythingv3, Orangemix, Counterfeit) users, make sure add '--no-half-vae' in command line of webui to prevent generating black images. See voldy's log for the details of edit method. For high-performance GPU, make sure 'medvram' and 'lowvram' mode are disabled.

This is a locon, please make sure you download the extension

recommended Weight: 0.6-1

recommended model: Anything4.5

Refer to https://civitai.com/models/23476/the-idolmster-cinderella-girls-starlight-stage-style-90-characters, thanks to the author

The main purpose of making this collection is to let the characters wear clothes that they don't usually wear, just like some characters in the example picture, you can try to replace the promt of the example picture with the character you want

Therefore, some characters may not be highly restored or have insufficient details, and have not learned specific clothes, so it is recommended to download or make lora of individual characters

I didn't do too many tests on each character because of the large amount of work, so I hope to get your review and feedback

You can also use this lora as a Artstyle lora, I used genshin character in the example

Please do not use this lora for any commercial purpose

参考了https://civitai.com/models/23476/the-idolmster-cinderella-girls-starlight-stage-style-90-characters， 感谢作者

The main purpose of making this collection is to let the characters wear clothes that they don't usually wear, just like some characters in the example picture, you can try to replace the promt of the example picture with the character you want

Therefore, some characters may not be highly restored or have insufficient details, and have not learned specific clothes, so it is recommended to download or make lora of individual characters

I didn't do too many tests on each character because of the large amount of work, so I hope to get your feedback

You can also use this locon as a painting style locon, in the example picture I used the character of the original god

Please do not use this locon for any commercial purposes

Recommeneded promt:

sparkle

（if you want the kirakira effect like game cards）

Characters promt:

Subaru Akehoshi, 1boy, male focus, blue eyes, solo, orange hair

Hokuto Hidaka, 1boy, male focus, blue eyes, solo, black hair

Makoto Yuuki, 1boy, male focus, green eyes, glasses, blonde hair

Mao Isara, 1boy, male focus, green eyes, red hair

Chiaki Morisawa, 1boy, male focus, solo, brown hair, brown eyes, orange eyes

Tetora Nagumo, 1boy, male focus, solo, multicolored hair, black hair, streaked hair, red hair, solo, smile, red eyes, teeth,

Kanata Shinkai, male focus, 1boy, green eyes, aqua hair, blue hair

Midori Takamine, 1boy, male focus, brown hair, blue eyes

Shinobu Sengoku, 1boy, male focus, hair over one eye, solo, purple hair, yellow eyes, multicolored hair, streaked hair

Rei Sakuma, 1boy, male focus, solo, red eyes, black hair

Kaoru Hakaze, 1boy, male focus, blonde hair, brown eyes

Koga Oogami, 1boy, male focus, solo, yellow eyes, grey hair

Adonis Otogari, 1boy, male focus, solo, purple hair, dark-skinned male, yellow eyes, dark skin

Eichi Tenshouin, 1boy, male focus, blonde hair,blue eyes

Wataru Hibiki, male focus, long hair, braid, blue hair, 1boy, purple eyes,

Tori Himemiya, 1boy, green eyes, male focus, solo, pink hair, sidelocks

Yuzuru Fushimi, 1boy, male focus, solo, blue hair, purple eyes

Nazuna Nito, male focus, 1boy, blonde hair, solo, red eyes

Tomoya Mashiro, 1boy, male focus, solo, blonde hair, brown eyes

Hajime Shino, male focus, 1boy, purple eyes, solo, blue hair, aqua hair

Mitsuru Tenma, 1boy, male focus, brown hair, brown eyes

Yuta Aoi, 1boy, male focus, green eyes, orange hair

Hinata Aoi, 1boy, male focus, green eyes, orange hair

Hiiro Amagi, 1boy, male focus, red hair, hat, blue eyes

Aira Shiratori, male focus, 1boy, blonde hair, solo, green eyes

Mayoi Ayase, 1boy, male focus, purple hair, solo, braid,long hair, ahoge

Tatsumi Kazehaya, 1boy, male focus, purple eyes, green hair

Tsukasa Suou, 1boy, male focus, purple eyes, solo, red hair

Izumi Sena, 1boy, male focus, blue eyes, grey hair

Leo Tsukinaga, 1boy, male focus, green eyes, solo, orange hair

Ritsu Sakuma, 1boy, male focus, red eyes, solo, black hair

Arashi Narukami, 1boy, male focus, blonde hair, blue eyes

Nagisa Ran, 1boy, male focus, solo, red eyes, long hair, white hair

Hiyori Tomoe, 1boy, green hair, purple eyes

Jun Sazanami, 1boy, male focus, blue hair, solo, yellow eyes

Ibara Saegusa, 1boy, male focus, red hair, blue eyes, solo, glasses

Rinne Amagi, 1boy, male focus, blue eyes, red hair, solo

HiMERU, 1boy, male focus, blue hair, yellow eyes

Niki Shiina, 1boy, male focus, blue eyes, blue hair, low ponytail

Kohaku Oukawa, 1boy, male focus, purple eyes, solo, pink hair

Kuro Kiryu, 1boy, male focus, solo, green eyes, multicolored hair, two-tone hair, black hair

Keito Hasumi, 1boy, male focus, glasses, green eyes, green hair

Souma Kanzaki, male focus, 1boy, solo, purple hair, purple eyes, long hair, ponytail

Mika Kagehira, 1boy, male focus, heterochromia, green hair, black hair, blue eyes, smile, jewelry, yellow eyes

Shu Itsuki, 1boy, male focus, pink hair, solo, blue eyes

Natsume Sakasaki, 1boy, red hair, solo, yellow eyes, multicolored hair, streaked hair

Tsumugi Aoba, 1boy, male focus, glasses, blue hair, yellow eyes, solo

Sora Harukawa, blonde hair, male focus, blue eyes, 1boy, green hair, multicolored hair, streaked hair

Madara Mikejima, 1boy, male focus, solo, brown hair, blue eyes

original description

A Character LoCon for Otori Akira (鳳アキラ) 【Helios Rising Heroes】

Weight:0.4～0.6

Hires. fix

trigger words: otori akira\(heliosrising_heroes\)

※以下日本語で失礼します。

運要素が絡みますがぽい衣装を出現させることは確認しています。

共通呪文:otoriakira\(heliosrising_heroes\), red hair, short hair, hair between eyes, bangs, green eyes, ear piercing, earrings, piercing, earclip,

制服

出現率アップ呪文:red shirt, black pants, white belt, police uniform, helios uniform,

ヒーロースーツ

出現率アップ呪文: open clothes, pants, black shirt, black pants, gloves, fingerless gloves, red gloves, hood, jacket, long sleeves, black gloves, open jacket, hood down, hooded jacket, red jacket, two-tone gloves, breastplate, the unyielding wild phoenix

※the unyielding wild phoenixはお守り呪文（効果薄）

水着（真夏のサバイバルアイランド）

出現率アップ呪文:sleeveless hoodie, blue hood, survival iland, red shorts, belt, sandals, bracelet, necklace, white hoodie

※再現を目指す場合、色指定箇所が多が多く混同しやすいのでガチャ覚悟必須。red shortsは丈が短くなる傾向のため期待薄ですがmale swimwear, swim trunksかpantsにして膝辺りから映さないようにするか…

IF世界線のアキラ（イクリプスアキラ）※原作にないので苦手な方はご注意を！

(white hair:1.7), (white jacket, black shirt, white pants:1.5), red eyes

ネガティブ:(red hair:1.6)

※数値は参考程度にお願いします。かなり頑固に赤髪と赤ジャケットを着たがる傾向にあるので数値は高く設定しました

Otori Akira - The Best of Helios Rising Heroes

Weight:0.4～0.6

Hires. fix

trigger words: otori akira\(heliosrising_heroes\)

*Please excuse me for the following in Japanese.

運要素が絡みますがPoi衣装を出現させることは確認しています。

共通呪文:otoriakira\(heliosrising_heroes\), red hair, short hair, hair between eyes, bangs, green eyes, ear piercing, earrings, piercing, earclip,

uniform

Appearance rate up spells: red shirt, black pants, white belt, police uniform, helios uniform,

hero suit

出現率アップ呪文: open clothes, pants, black shirt, black pants, gloves, fingerless gloves, red gloves, hood, jacket, long sleeves, black gloves, open jacket, hood down, hooded jacket, red jacket, two-tone gloves, breastplate, the unyielding wild phoenix

*The unyielding wild phoenix is ​​a charm spell (less effective)

Swimsuit (Midsummer Survival Island)

Appearance rate spells: sleeveless hoodie, blue hood, survival island, red shorts, belt, sandals, bracelet, necklace, white hoodie

* When aiming to reproduce, it is necessary to be prepared for gacha because there are many color designations and it is easy to get confused. Red shorts tend to be short, so I don't have much hope for them, but should I wear male swimwear, swim trunks or pants so that they don't show around the knees...

IF Sekaisen no Akira (Eclipse Akira)

(white hair:1.7), (white jacket, black shirt, white pants:1.5), red eyes

Negative: (red hair: 1.6)

*Please use the figures for reference only. She has a tendency to stubbornly want to wear red hair and a red jacket, so I set the numbers high.

A LoHa trained with 12 images. 1 batch size. Uses my own method.

If you want to support me now you can: https://ko-fi.com/truefalse

Trained 10 epochs, car looks great. lowered some settings to make the files smaller, think the quality is still there while the files are smaller. Great!

image grid: https://files.catbox.moe/k5ocq7.png

This LORA is Misawa Maho from RO-KYU-BU!

initial tags:

"misawa maho, twintails, fang"

main costume:

"gym uniform, bike shorts"

"basketball uniform, shorts"

You may want to use 7thLayerAbyss or AOM3A.

show you guts cool say what saikou daze.

A LoHa made using a 30 image dataset and my own methods. 15 batch size.

This was a test after being requested, it uses somebody else's dataset that was left untagged save for the triggerword. Uses my own method. 824 images 15 batch size.

If you'd like to support me and do more:
https://ko-fi.com/edgonkofi - commissions open!
If you're looking for a >>LoRA Making Tutorial<<

Urban Lingerie

We all like hoodies and lingerie.

Why not combine urban wear and lingerie then?

Get creative and mix prompts from my Urban Gal and KPG models!

This is a LoHa model trained from 123 various sources images. The tests found that portraits performed better,which could create a very dynamic expression,and that's where the name comes from.

You can use the LoHa to designeyes & face.

Weight 0.9-1.0 will perform better

original description

Of course the tirgger word is still 光头强lora

Support my video here:https://www.bilibili.com/video/av823626310

My QQ group:https://jq.qq.com/?_wv=1027&k=o89HEhaS

Of course the tirgger word is still bald strong lora

Support my video here:https://www.bilibili.com/video/av823626310

My QQ group:https://jq.qq.com/?_wv=1027&k=o89HEhaS

It's a loha model trained wtih Pixiv painter opinew6.

The cute and vivid style always hits my heart straight.

So I'd like to share it with you guys all.

It works well when mixed with other Loras, so you can just create different stylized images of your own Lora.

Weight from 0.6 to 1.0 works ok.

Currently, the weight can be from 0.6 to 1.0.

You should refer it in prompt words instead of Additional Network.

Note that the loha model cannot be used in Additional Network, you should quote it in the prompt.

这是一个画风LoCon，

This is an artstyle LoCon,

It seems not very stable at the moment, when I learn new training methods, I will update this LoCon.

It doesn't seem very stable at the moment, I will update this LoCon once I learn new training methods.

Note: The trigger word is: chibi,

Note: The trigger phrase is: chibi,

It is recommended to use it with the character lora.

I recommend using it with a character Lora.

It is recommended to set the intensity of this LoCon to 1, the intensity of the character lora to 0.8, set the CFG to 7, and use anything-v3 for image generation.

Setting the weight of this LoCon to 1 and the weight of the character Lora to 0.8，set CFG to 7,use https://civitai.com/models/66/anything-v3 。

Adjust according to the actual situation to get a better picture effect.

Make adjustments according to the actual situation to achieve better visual effects.

If you have any experience related to painting style training, please share with me.

If you have any insights on artstyle training, please feel free to share them with me,

so that I can modify this LoCon.

so that I can modify this LoCon.

• Weight: 1.0 - 1.1

• CFG Scale: 5.5 - 7

• Recommended Prompts: manaka nemu, brown hair, purple eyes, blunt bangs, school uniform, neckerchief, hair bun, sidelocks, long hair

original description

Buy us a cup of coffee and we’ll spill the beans.

https://www.buymeacoffee.com/CoSubcreation

Or Fanbox

https://co-subcreation.fanbox.cc/

Or爱发电

https://afdian.net/a/Co_Sub_creation

 

Your support is the fuel that powers our world-building. Your ideas are the bricks that shape our world-making!

あなたの応援で、世界は生まれ変わる！あなたのアイデアが、世界を作り上げる原動力になるよ！

您的支持将成为创造世界的动力，您的想法将成为构建世界的基石！

 

 

If you have any ideas, suggestions, or questions during use, you can ask them on Discord ↓

ご利用中に何かアイデアや提案、質問がございましたら、Discordにてお問い合わせいただけますようお願いいたします↓

使用中有任何想法&建议&问题，可以去Discord提问↓

https://discord.gg/tVpJMhjUwx

 

 

If you’re new to LoHa, please read this:

This model was trained with LoHa, so you need to install the WebUI extension. Trust me, it’s worth it.😉

https://github.com/KohakuBlueleaf/LyCORIS

LoHa model path: models/Lora

Usage: Click on the 🎴 button under “Generate” in the WebUI, choose the Lora tab, select the model, and adjust the weights. It’s as easy as pie.🍰

第一次使用LoHa请看：

本模型训练使用了LoHa，需要安装WebUI的扩展。

https://github.com/KohakuBlueleaf/LyCORIS

LoHa模型路径: models/Lora

使用：点击WebUI“生成”下面的🎴，选择Lora 选项卡，选择模型，更改权重。

LoHaを初めて使う場合は、こちらをご覧ください：

このモデルトレーニングにはLoHaが使用されており、WebUIの拡張機能をインストールする必要があります。

https://github.com/KohakuBlueleaf/LyCORIS

LoHaモデルのパス：models/Lora

使い方：WebUIの「生成」の下の🎴をクリックし、Loraタブを選択して、モデルを選択して、重みを変更してください。

 

 

Recommended parameters:

推奨パラメータ：

推荐参数：

Model: blueboys2D_v10

Vae: kl-f8-anime2

Clip skip: 2

weight:0.7~0.9

Sampler: DPM++ 2M Karras

Steps: 25

Size: 512x512|384x640|640x384

CFG Scale: 9 or 14~17(see Instructions: need extension)

Extension setting↓

Enable Dynamic Thresholding (CFG Scale Fix)

Mimic CFG Scale:7

Advanced Options▼

Top percentile of latents to clamp:100

Mimic Scale Scheduler:Constant

Minimum value of the Mimic Scale Scheduler:0

CFG Scale Scheduler:Half Cosine Up

Minimum value of the CFG Scale Scheduler:3

 

 

Preferred Prompts:

推奨TAG：

推荐提示词：

Prompt:

1boy,solo,upper body,fata, <lora:Style_GlassyandGlaze (FATA)_v1.0:1:1,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,1>(see Instructions)

Negative Prompt:

EasyNegative, badhandv4

EasyNegative | Stable Diffusion TextualInversion | Civitai

badhandv4 - AnimeIllustDiffusion | Stable Diffusion TextualInversion | Civitai

 

 

Instructions:

Pulling the CFG too high can lead to flower patterns on your artwork, but fear not! You can install the Dynamic Thresholding (CFG Scale Fix) plugin to fix the issue:

https://github.com/mcmonkeyprojects/sd-dynamic-thresholding

We recommend increasing the weight, but be warned, it may lead to overfitting. We suggest using the Lora block weight extension to adjust the layering (you can use the recommended values directly from the prompt):

https://github.com/hako-mikan/sd-webui-lora-block-weight

It works well with some Sairu and Hira-nuri base models, but since the model was trained on Blueboys2D, we recommend sticking with the original base model for the best results. It does not support pastel colors, thick paint styles, 2.5D, or 3D base models.

It's great for drawing upper bodies and faces, but drawing full-body images may cause distorted faces, so you'll need to import the image into the image-to-image local redraw tool.

The default hair color is black and the hairstyle is an M-shape. To change the hair color, hairstyle, or eye color, you'll need to add specific prompts (as shown in the example image).

I don't use high-definition restoration because running multiple HD images at once during the lottery stage can't guarantee high-quality results. Instead, I prefer to batch process several good-quality images using the txt2img tool (usually with good composition), then use img2img to increase the resolution of individual images and determine the level of detail to add by adjusting the denoising strength. Finally, for those who really need HD images, I'll use Extras to generate high-definition restoration images at the end.

説明：

CFGの値を高くしすぎると、花模様が発生することがあります。この問題を修正するために、Dynamic Thresholding (CFG Scale Fix)プラグインをインストールすることをお勧めします。

https://github.com/mcmonkeyprojects/sd-dynamic-thresholding

重みは少し高めに設定することをお勧めしますが、オーバーフィッティングが発生する可能性があります。分層を調整するには、lora block weight拡張機能を使用することをお勧めします（推奨ヒントの数値を直接使用できます）。

https://github.com/hako-mikan/sd-webui-lora-block-weight

セルルロース、平塗りタイプの背景をサポートしていますが、学習モデルの背景はblueboys2Dであり、より良い効果を得るためには、元の背景を使用することをお勧めします。また、粉彩系、厚塗り系、2.5D、3Dタイプの背景はサポートされていません。

顔や上半身の描写には慣れていますが、全身の描写は難しいため、一部分を描いた後、txt2imgで画像を再構築する必要があります。

デフォルトの髪色は黒色で、髪型はM字型です。髪色、髪型、瞳の色を変更したい場合は、適切な指示語を追加する必要があります（例えば、図のように）。

抽選フェーズで多数の画像を一度に生成すると、高品質の画像を保証することができません。通常、私は画像の品質が良好で構図が適切な画像を一括生成し、img2imgを使用して単一の画像の解像度を高く調整し、Denoising strengthを調整して詳細度を調整します。最後に、Extrasで高品質の修正画像を生成することができます。

说明：

CFG拉的过高会导致花图，可以安装插件Dynamic Thresholding (CFG Scale Fix)来修复：

https://github.com/mcmonkeyprojects/sd-dynamic-thresholding

权重建议高一点，但可能会导致过拟合，建议使用扩展lora block weight来调整分层（可以直接使用推荐提示词里的数值）。

https://github.com/hako-mikan/sd-webui-lora-block-weight

这里是大佬的一些关于这个插件的教程和预设@大江户战士

https://www.bilibili.com/video/BV1Wv4y157NH/

支持一些赛璐璐、平涂类底模，但由于训练的底模就是blueboys2D，所以如果您想得到较好的效果，还是建议使用原底模。不支持粉彩系、厚涂系、2.5d、三次元底模。

擅长画上半身和脸，画全身容易崩脸，需要导进图to图局部重绘。

默认发色是黑色，发型是m形，想要更改发色、发型和瞳色需要加提示词（如例图）。

我没有使用高清修复是因为，在抽卡阶段一次性跑多张无法保证画面质量的高清图比较浪费时间，通常我比较喜欢在txt2img批量跑出还不错的图后（通常是构图比较好），转img2img直接拉高分辨率跑单张，并通过调整Denoising strength来决定添加细节的程度，最终跑出符合预期的图。如果确实需要高清图，在最后才会拉到Extras生成高清修复图。

 

In fact, you can also use img2img to generate beautiful anime girls, for example: ↓

実は、img2imgを使って美しいアニメの女の子を生成することもできます。例えば、↓

其实它也可以用img2img来生成好看的妹子，比如↓

Buy us a cup of coffee and we’ll spill the beans.

https://www.buymeacoffee.com/CoSubcreation

Or Fanbox

https://co-subcreation.fanbox.cc/

Or Love Power Generation

https://afdian.net/a/Co_Sub_creation

Your support is the fuel that powers our world-building. Your ideas are the bricks that shape our world-making!

With your support, the world will be reborn! Your ideas will be the driving force that creates the world!

Your support will become the motivation to create the world, and your ideas will become the cornerstone of the world!

If you have any ideas, suggestions, or questions during use, you can ask them on Discord ↓

If you have any ideas, suggestions, or questions while using it, please contact us on Discord ↓

If you have any ideas & suggestions & questions during use, you can go to Discord to ask questions↓

https://discord.gg/tVpJMhjUwx

If you’re new to LoHa, please read this:

This model was trained with LoHa, so you need to install the WebUI extension. Trust me, it’s worth it.😉

https://github.com/KohakuBlueleaf/LyCORIS

LoHa model path: models/Lora

Usage: Click on the 🎴 button under “Generate” in the WebUI, choose the Lora tab, select the model, and adjust the weights. It’s as easy as pie.🍰

For the first time to use LoHa, please see:

This model training uses LoHa, and the extension of WebUI needs to be installed.

https://github.com/KohakuBlueleaf/LyCORIS

LoHa model path: models/Lora

Use: Click 🎴 under "Generate" in the WebUI, select the Lora tab, select the model, and change the weight.

If you are new to LoHa, please see here:

This model training uses LoHa and requires the WebUI extension to be installed.

https://github.com/KohakuBlueleaf/LyCORIS

LoHa model path: models/Lora

How to use: Click 🎴 under "Generate" in the WebUI, select the Lora tab, select the model, and change the weights.

Recommended parameters:

Recommended parameters:

Recommended parameters:

Model: blueboys2D_v10

Vae: kl-f8-anime2

Clip skip: 2

weight:0.7~0.9

Sampler: DPM++ 2M Karras

Steps: 25

Size: 512x512|384x640|640x384

CFG Scale: 9 or 14~17(see Instructions: need extension)

Extension setting↓

Enable Dynamic Thresholding (CFG Scale Fix)

Mimic CFG Scale:7

Advanced Options▼

Top percentile of latents to clamp:100

Mimic Scale Scheduler:Constant

Minimum value of the Mimic Scale Scheduler:0

CFG Scale Scheduler:Half Cosine Up

Minimum value of the CFG Scale Scheduler:3

Preferred Prompts:

Recommended TAG:

Recommended prompt words:

Prompt:

1boy,solo,upper body,fata, (see Instructions)

Negative Prompt:

EasyNegative, badhandv4

EasyNegative | Stable Diffusion TextualInversion | Civitai

badhandv4 - AnimeIllustDiffusion | Stable Diffusion TextualInversion | Civitai

Instructions:

Pulling the CFG too high can lead to flower patterns on your artwork, but fear not! You can install the Dynamic Thresholding (CFG Scale Fix) plugin to fix the issue:

https://github.com/mcmonkeyprojects/sd-dynamic-thresholding

We recommend increasing the weight, but be warned, it may lead to overfitting. We suggest using the Lora block weight extension to adjust the layering (you can use the recommended values directly from the prompt):

https://github.com/hako-mikan/sd-webui-lora-block-weight

It works well with some Sairu and Hira-nuri base models, but since the model was trained on Blueboys2D, we recommend sticking with the original base model for the best results. It does not support pastel colors, thick paint styles, 2.5D, or 3D base models.

It's great for drawing upper bodies and faces, but drawing full-body images may cause distorted faces, so you'll need to import the image into the image-to-image local redraw tool.

The default hair color is black and the hairstyle is an M-shape. To change the hair color, hairstyle, or eye color, you'll need to add specific prompts (as shown in the example image).

I don't use high-definition restoration because running multiple HD images at once during the lottery stage can't guarantee high-quality results. Instead, I prefer to batch process several good-quality images using the txt2img tool (usually with good composition), then use img2img to increase the resolution of individual images and determine the level of detail to add by adjusting the denoising strength. Finally, for those who really need HD images, I'll use Extras to generate high-definition restoration images at the end.

explanation:

If the CFG value is too high, flower patterns may occur. I recommend installing the Dynamic Thresholding (CFG Scale Fix) plugin to fix this issue.

https://github.com/mcmonkeyprojects/sd-dynamic-thresholding

Setting the weights slightly higher is recommended, but overfitting can occur. To adjust the division layers, I recommend using the lora block weight extension (you can use the suggested hint numbers directly).

https://github.com/hako-mikan/sd-webui-lora-block-weight

It supports cellulose, plain type background, but the learning model background is blueboys2D, and for better effect, it is recommended to use the original background. In addition, powder, thick, 2.5D, and 3D type backgrounds are not supported.

I'm used to drawing the face and upper body, but drawing the whole body is difficult, so after drawing a part, I need to reconstruct the image with txt2img.

The default hair color is black and the hairstyle is M-shaped. If you want to change the hair color, hairstyle, or eye color, you need to add the appropriate directives (for example, as shown).

Generating a large number of images at once during the drawing phase cannot guarantee high quality images. Usually I batch generate images with good image quality and good composition, then use img2img to adjust single image resolution higher and denoising strength to adjust detail. Finally, Extras can generate high-quality corrected images.

illustrate:

If the CFG is pulled too high, it will cause a flower pattern. You can install the plug-in Dynamic Thresholding (CFG Scale Fix) to fix it:

https://github.com/mcmonkeyprojects/sd-dynamic-thresholding

The weight is recommended to be higher, but it may lead to overfitting. It is recommended to use the extended lora block weight to adjust the layering (you can directly use the value in the recommended prompt word).

https://github.com/hako-mikan/sd-webui-lora-block-weight

Here are some tutorials and presets from Big Guy about this plugin @大画DO士人

https://www.bilibili.com/video/BV1Wv4y157NH/

Some bases of celluloid and flat coating are supported, but since the training base is blueboys2D, it is recommended to use the original base if you want to get better results. Pastel, impasto, 2.5d, and three-dimensional bottom molds are not supported.

Good at drawing upper body and face, drawing the whole body is easy to collapse the face, need to import map to map and redraw partly.

The default hair color is black, and the hairstyle is m-shaped. If you want to change the hair color, hairstyle and eye color, you need to add prompt words (such as the example picture).

I didn’t use HD repair because it was a waste of time to run multiple high-definition images that cannot guarantee the picture quality in the card drawing stage. Usually, I prefer to run txt2img batches of good images (usually the composition is better), Turn to img2img to directly pull the high-resolution run sheet, and adjust the Denoising strength to determine the degree of adding details, and finally run out a picture that meets expectations. If a high-definition image is really needed, it will be pulled to Extras to generate a high-definition restoration image at the end.

In fact, you can also use img2img to generate beautiful anime girls, for example: ↓

Actually, you can also use img2img to generate beautiful anime girls. For example, ↓

In fact, it can also use img2img to generate beautiful girls, such as ↓

This is a costume LoHa. Not character LoHa.

weight: 0.6-0.7

Walkure romance style

Can generate knight armor with pleated skirt and school uniform of Walkure Romanze.

No trigger words are set for the characters.

Samples are generated by t2i -> i2i -> Tiled Diffusion+Tiled VAE. Some prompts were deleted by Tiled Diffusion+Tiled VAE process, so the prompt of the samples is incomplete.

Samples full prompt:

1. (masterpiece, best quality,extreme detailed beautiful eyes), best illumination, perfect lighting,look at viewer, (1 girl, solo,street, upper body,dynamic angle),(victory pose:1.2), (smile, shiny_eyes,large breasts),sidelocks,long hair, armor, (white hair:1.5), (ponytail,golden eyes:1.4),breastplate, red skirt, thighhighs, greaves, gauntlets, lace_trim,

2. (masterpiece, best quality,extreme detailed beautiful eyes), best illumination, perfect lighting,look at viewer, (1 girl, solo,street, upper body,dynamic angle),(spread legs:1.2), (smile, shiny_eyes,large breasts),sidelocks,long hair, , (white hair:1.4),(ponytail:1.5), (golden eyes),white competition swimsuit, thighhighs,black greaves,black gauntlets, lace_trim,

3. (masterpiece, best quality,extreme detailed beautiful eyes), best illumination, perfect lighting,look at viewer, (1 girl, solo,bedroom, upper body,dynamic angle),(spread legs:1.2), (smile, shiny_eyes,large breasts),sidelocks,long hair, (white hair:1.4),(ponytail:1.5), (golden eyes:1.3),military_uniform, thighhighs, aiguillette, epaulettes,pleated_skirt, zettai_ryouiki,panties,

This is retrained fromthesauceerdataset, just a version of hisTutu LoRA, please check out his original work! He was kind enough to share his dataset.

This is a LoHa LoRA! You need the extension!
Why? LoHa have virtually no effect in output art style and they do a lot better with multi-concepts.

Trigger wordsfor V2.20:

positive: (tutu)

You can also try "ribbon-trimmed skirt", "layered skirt" and "miniskirt".
If failing to make full shirt, try adding either "leotard" or "tank top" or "covered navel" to prompt.

Keep in mind you might have to adjust LoRA and/or tag weight for it to appear, defaults to "ribbon-trimmed skirt" most of the time! Dataset wasn't an exactly split between the different concepts.

Follow the steps in the Model Version notes to reproduce my images.

Sample images generated with Hassaku.

All negative embeds should be easy to find. and Prompts should be included in the sample images! Click the little icon in the bottom right to see it.

It was trained on 62 512x512 images provided by thesauceer, 20 epochsusing Animefull-final-prunedfor the Training Model.

Danbooru tags generated by Waifu Diffusion 1.4 Tagger.

LyCORIS of Jaroslaw Kaczynski

LyCORIS of Janusz Korwin-Mikke

Weights: 0.8-1.4

This is an experimental and somewhat unstable attempt at squeezing 9 different versions of Shinobu Oshino/Kiss-shot from the Monogatari series into one model. Getting desired results may be difficult, please check the preview images for the most reliable prompts for each version of the character!

Versions and tokens:

• Adult Kiss-shot | kiss-shot acerola-orion heart-under-blade

• Teen Kiss-shot (Kizumonogatari) | kizu_teen

• Teen Kiss-shot (TV) | tv_teen

• White dress (Kizumonogatari) | kizu_long

• Pink dress/short hair (Kizumonogatari) |

• White dress (TV) | tv_white

• Pink dress (TV) | tv_pink

• School uniform (TV) | tv_student

• Gothic dress (TV) | tv_gothic

Again, check the preview images for example prompts!

LyCORIS of our boy Asmongold.

Suggested weights: 0.8 - 1.4

LyCORIS of Antony Starr as Homelander.

This was trained on purely Homelander pictures by design. Still, it is sometimes possible to get other characters.

Suggested weights: 0.8 - 1.4

LoCon trained on the artstyle of kojima ayami, who did the art for a lot of Castlevania games.

It is not quite her style but still looks good enough on it's own.

Trigger word is SotnStyle.

I recommend weight 0.7~0.8, or 0.9~1 for higher resolutions.

Trained on NAI.

LyCORIS of Thomas Jane

Suggested weights: 0.8 - 1.4

A Character LoCon for Niki Shiina [Ensemble Stars!]

Weight:0.4～0.7

Hires. fix

trigger words: shiina niki\(ensemble, stars!\)

*Please excuse me for the following in Japanese.

運要素が絡みますがPoi衣装を出現させることは確認しています。

Common spells: shiina niki\(ensemble, stars!\), long hair, bangs, blue eyes, shirt, gray hair, hair between eyes, ponytail

Crazy:B costume

Appearance increase spells: jacket, choker, belt, necklace, black pants, open jacket, white shirt, white jacket, Crazy:B jacket, Crazy:B Uniform

hot limit costume

Appearance rate spells: navel, jacket, purple pants, necklace, ring, tank top, yellow jacket, open jacket, open clothes, hot limit_hot summer

Hot Limit Outfit (SCR)

Appearance rate up spells: navel, jacket, purple pants, necklace, ring, tank top, purple jacket, open jacket, open clothes, hot limit_hot summer

Please see https://huggingface.co/alea31415/bofuri-full for details

It seems that many people are not aware of my bofuri model so I extract a locon as usual for easier use.

The model is trained on top of ACertainty with clip skip 1.

Trigger words

• BoDrag

• BoDread

• BoFrederica

• BoIz

• Canadian

• Bo Kasumi

• Kasumi Sarashi

• Bo Kuromu

• BoMaple

• BoMaple angel form

• BoMaple beginner outfit

• BoMaple black armor

• BoMaple machinegod form

• BoMaple pajama

• BoMaple sheep form

• BoMaple uniform

• BoMarcus

• BoMay

• BoMii

• BoMisery

• BoPain

• BoSally

• BoSally cape

• BoSally pajama

• BoSally turtleneck scarf

• BoSally uniform

• Head

• BoYui

• dorazou

• Fox Oboro

• Monster Atrocity

• Turtle Syrup

• Hydra

Anything + dataset(FineTune)

I've been thinking about things lately, you know, like the things we think about when we're not really thinking about anything.

This is still an early version and will be improved in the future.

Enjoy!

Contact → My Twitter

Note:This is a Locon. Install the extension to get it to work. I've found them to do better when working on ideas that I want easily added to one of my LORA characters. Also, they're WAY less damaging to images, even at high weights.

I had made a couple NSFW style Locons (and I have a few more in development with sneak peaks here), but I don't always want my waifu's to become horny and evil. So. Here I am combating the evil part with an Angel (warning, still can be horny though). It was trained on animefull_final with 29 images with the usual self-pruning and tagging and a wide range of different images. I like keeping my prompts simple for the first few pictures so that you can see it without a whole bunch of mumbo jumbo. Make your pictures better than mine though!

What does it do you ask? A Locon that will transform your waifu into an angelic form. Usually with wings. I leave these Locon's very general, so if you want to push a specific direction with your images (like a halo/wings), I'd tag it. Otherwise it'll be random on what form it takes.

How to use it: I've found that with these Locon's I've been submitting, weight's can be wide and varied. 0.5 is fine is you want only a small effect. But a lot of time you can go as high as 1.5 if you want to really push the LORA you're changing.

Also the trigger word is:

AngelT -Only trigger word currently added.

Optional - Of course you can add other words that describe angels if you want specific components and to make sure they get into the picture

Extra, Extra!: Should be well trained and usable in most models. Images shown for example.

Picture Descriptions:

1-3: AnythingV3 using just this Lora. Showing different types of outfits can be called when just using this.

4: AnythingV3 combo'd with my Seraphina (Disgaea 5) Lora.

5: AnythingV3 combo'd with my in progress Ghislaine (Mushoku Tensei) Lora.

6: AnythingV3 combo'd with my in progress V2Corruption Locon.

7: AnythingV3 combo'd with my in progress Brainwashing Locon.

8: AnythingV3 combo'd with my in progress "Symbioted Taking Over" Locon (basically a symbiote merging with the user).

9: HassakuHentaiModel just with the Lora.

10: AbyssOrange with just the Lora.

11: PerfectWorld using just this Lora.

What's next for this? Unsure. Feedback will be the decider if I do a V2.

What's next that isn't this?

More character Loras. (Ghislaine from Mushoku Tensei, Conqueror Lilias from E7, Rozalin from Disgaea 2, Tharja from Fire Emblem, Catherine from Catherine Full Body, Miku + Piscalat from Idaten Deities, Kuribayashi Shino from GATE, Rin Sohma from Fruits Basket, "Kasumi" from P5, Ann from P5, Ikumi Mito from Food Wars, Alice Nakiri from Food Wars, Chika Fujiwara from Kaguya-sama, Lulu from FF10, Momo from My Hero, Vivy from Vivy, Dominique de Sade from Vanitas, and Dos from One Punch Man.) And maybe Wilhelm van Astrea, my favorite Re:Zero husbando or a Prinny from Disgaea.

V2's of some of the previous Loras. Specifically make them more diversely acceptable with other models and LORA's.

I also have some more quite NSFW Locon's. Specifically, being pet play, being stepped on, brainwashing, corruptedv2, and one that might be too much for me to even upload here >.<.

And, uh, I want to make a few "normal" Locon's so I have an "knight-ify" your character one coming up. And a forest fairy one as well. And a soldier one. And a barbarian one.

Or just give me some random suggestion or priority opinions. That works too.

Two versions, soft and hard.

enjoy yourself!

• Welcome to the AI ​​Painting Model Museum QQ group

< li>

①Group: 791222249

• ②Group: 599962550

Locon trained on 81 cum filled shoes.

Performed well when drawing shoes only.

Have a great tendency of making any shoes into high heels.

A Locon of ba-zerarudo from Frame Arm Girls.

Trained Trigger words:

ba-zerarudo | leotard | pantyhose | rabbit tail | high heels

Pictures below are generated on Any4.5 with Controlnet.

BOOOOBA Eldridge :D

Trained trigger words:

Eldridge

default outfit | Kitty idol | Holy Night's Embrace | The Inquisitive Recluse | New Year's Outing

The "New Year's Outing" is overfitted and may have poor performance.

I trained this LyCORIS by collecting about 200 images generated by NijiJourney V5 on the internet.

Set weights to 0.7~0.8. I generated the example images with AOM3.

loha training just for fun

Hires step 20

Denoising 0.6

This model was made trained on 160 images made by Chelodoy, most of theme were NSFW, but the model works flawlessly with any type of image, it is black magic.

Trigger word is chelodoystyle.But it is unnecessary to use it. Model should work just fine at 1 weight.

A Character LoCon for rinne amagi [Ensemble Stars!]

Weight:0.4～0.7

Hires. fix

trigger words: amagi rinne\(ensemble stars!\)

*Please excuse me for the following in Japanese.

運要素が絡みますがPoi衣装を出現させることは確認しています。

Common spells: amagi rinne\(ensemble stars!\), red hair, short hair, blue eyes, hair between eyes, jewelry, earrings, ring

*Hair between eyes gets in the way when trying to recreate plain clothes

Crazy:B costume

Appearance increase spells: jewelry, jacket, black pants, open jacket, chain, ring, white shirt, white jacket, Crazy:B jacket, Crazy:B Uniform

plain clothes (headband)

Appearance rate up spells: headband, black shirt, pants

Lucky Chain (Feature Scout Rinne Costume)

Appearance rate spells: jacket, necklace, off shoulder, tank top, torn clothes, fur trim, black shirt, black pants, ring, fur-trimmed jacket, red belt, chain necklace

Flambé! costume

出現率アップ呪文:gloves, hat, jewelry, jacket, earrings, black gloves, pants, fingerless gloves, collar, black jacket, torn pants, black pants, ring, peaked cap, single glove, torn pants, military cap

A dream style LoRA for Stable Diffusion webui.
一个梦幻风格的LoRA。

Girl in the Flask (Girl in the Flask)

An Utility/Artstyle LoCon

You can use this Locon model to bottle up the characters you want and take them with you.

You can use this locon model which will allow you to bottle the character you want to take away.

• Recommeneded prompts: simple background, full body, air bubble, water,submerged (Please check my sample work for more details)

• Trigger word: in bottle,in container

• You can use this LoCon with other characters' LoRA.

• Recommeneded Weight: 0.6-1 (You can adjust it by your perference.)

• CFG scale: 5.5-7

• Check point recommendation: Anything V5, Dan mix

• This LoCon may have some small issues, I will fix it soon.

• How to use LoCon/LyCORIS? Please read instruction of a1111-sd-webui-locon.

Human kind can not gain anything without first giving something in return. To obtain something of equal value must be lost. - Fullmetal Alchemist

If you like my work,please give me a like <3.

If you like my model please give me a like <3.

If you like my model, please click like <3.

original description

::: 该页面为中国传统纹样loha的使用介绍，关于详细的训练过程，分层调试方法，可以到我们的wiki上去查看。https://aigc.ioclab.com/sd-showcase/chinese-ornament.html :::

中国传统纹样lora by IoC Lab

训练及数据：@shichen @ciaochaos

版本 介绍 2023.4.1 上传lora，介绍内容（本篇） 2023.4.4 更新训练方法（wiki待上传）

在图片中你可以看到有大约25个以上的版本，我挑选了其中最有特点的几个版本，你可以在version中挑选下载，或者前往huggingface下载全部的版本

https://huggingface.co/ioclab/ioc-lab-models/tree/main

使用方法：

在训练集的标注中，分为以下五个语段：

traditional chinese pattern

with circular symmetrical composition
with square symmetrical composition
with special-shaped symmetrical composition

red and blue color scheme

featuring horses and clouds
featuring flowers and branches

a flower in the center

所以你可以围绕着这五种固定句式进行扩充，如需要作为背景，可以调低权重，增加 1 girl

分层拆解：

在分层图片中可见，在高权重下，通过仅开启 midd，outd 等 lora 层的权重，可以做到相比于直接降低权重，更能保留纹样风格的人物背景图。

推荐模型版本介绍：

这里是一些效果比较特色的模型版本，但出图效果与参数，种子等都有关系，你可以尽情尝试，期待你的评论

特色模型版本

以下是一些特色模型版本：

• 4：特点是纹样更卡通化，色彩更缓和

• 7：特点是对关键词的还原更好，更符合数据集的倾向

• 10：特点是更对称，同时构图倾向会偏向于非中心对称构图

• 13：特点是相比10的颜色会更缓和，高权重下会出现第三种散布构图

• 15：特点是高饱和度和高横向对称构图

• 16：特点是高饱和度+散布构图

人物配合版本

以下是容易和人物配合的版本：

• loha01_old3: 更容易出卡通光影，构图和花纹装饰性强

• loha04_old3: 构图容易出全身，光影保留效果好

• loha06_old1: 对构图限制小，人物动作自然

• loha02_old2: 背景层次更多，对称性强

数据集标注及loha训练方法，见wiki链接

https://aigc.ioclab.com/sd-showcase/chinese-ornament.html

Chinese Ornament Wiki in English

::: This page is an introduction to the use of traditional Chinese pattern Lora, for detailed training processes and layer debugging methods, please refer to our wiki. https://aigc.ioclab.com/sd-showcase/chinese-ornament.html :::

Chinese traditional pattern Lora by IoC Lab

Training and Data: @shichen @ciaochaos

Version Description 2023.4.1 Uploaded Lora, Introduction Content (This Article) 2023.4.4 Updated Training Method (Wiki To Be Uploaded)

In the pictures, you can see more than 25 versions. I selected the most distinctive versions for you to download, or you can go to Huggingface to download all versions.

ioclab/IocLab_Models at main (huggingface.co)

How to use:

In the annotation of the training set, the following five language segments are divided:

traditional chinese pattern

with circular symmetrical composition
with square symmetrical composition
with special-shaped symmetrical composition

red and blue color scheme

featuring horses and clouds
featuring flowers and branches

a flower in the center

So you can expand around these five fixed sentence patterns. If you need to use it as a background, you can reduce the weight and add one girl.

Layered Decomposition:

As shown in the layered images, under high weight, by only turning on the weights of midd, outd, and other Lora layers, you can achieve a person-background image with a style more retained than directly reducing the weight.

Recommended Model Version Introduction:

Here are some model versions with distinctive effects, but the output effect is related to parameters, seeds, etc. You can try them out and we look forward to your comments.

Distinctive Model Versions

Here are some distinctive model versions:

• 4: The characteristic is that the pattern is more cartoonish and the color is more gentle.

• 7: The characteristic is that it better restores the keywords and is more in line with the tendency of the dataset.

• 10: The characteristic is that it is more symmetrical, and the composition tends to be non-centrally symmetrical.

• 13: The characteristic is that compared to 10, the color will be more gentle, and the third scattered composition will appear under high weight.

• 15: The characteristic is high saturation and high horizontal symmetry composition.

• 16: The characteristic is high saturation + scattered composition.

Character Matching Versions

The following are versions that are easy to match with characters:

• loha01_old3: More likely to produce cartoon light and shadow, with strong composition and decorative patterns.

• loha04_old3: The composition is easy to produce the whole body, and the effect of retaining light and shadow is good.

• loha06_old1: The restriction on composition is small, and the character's movements are natural.

• loha02_old2: More layers in the background, strong symmetry.

For dataset annotation and Lora training methods, please refer to the wiki link.

https://aigc.ioclab.com/sd-showcase/chinese-ornament.html

::: 该页面为中国传统纹样loha的使用介绍，关于详细的训练过程，分层调试方法，可以到我们的wiki上去查看。https://aigc.ioclab.com/sd-showcase/chinese-ornament.html :::

Traditional Chinese pattern lora by IoC Lab

Training and data: @shichen @ciaochaos

Version Introduction 2023.4.1 upload lora, introduction content (this article) 2023.4.4 update training method (wiki to be uploaded)

In the picture, you can see that there are more than 25 versions. I have selected the most characteristic versions. You can choose to download in version, or go to huggingface to download all versions.

https://huggingface.co/ioclab/ioc-lab-models/tree/main

Instructions:

In the annotation of the training set, it is divided into the following five segments:

traditional chinese pattern

with circular symmetrical composition
with square symmetrical composition
with special-shaped symmetrical composition

red and blue color scheme

featuring horses and clouds
featuring flowers and branches

a flower in the center

So you can expand around these five fixed sentence patterns. If you need to use it as a background, you can lower the weight and add 1 girl

Layered disassembly:

It can be seen in the layered pictures that under high weight, by only turning on the weight of midd, outd and other lora layers, it is possible to achieve a character background image that can better retain the pattern style compared to directly reducing the weight.

Recommended model version introduction:

这里是一些效果比较特色的模型版本，But the effect of drawing is related to parameters, seeds, etc., you can try it as much as you want, and look forward to your comments

featured model version

以下是一些featured model version：

• 4：特点是纹样更卡通化，色彩更缓和

• 7：特点是对关键词的还原更好，更符合数据集的倾向

• 10：特点是更对称，同时构图倾向会偏向于非中心对称构图

• 13：特点是相比10的颜色会更缓和，高权重下会出现第三种散布构图

• 15：特点是高饱和度和高横向对称构图

• 16：特点是高饱和度+散布构图

Character matching version

Here's a version that's easy to work with characters:

• loha01_old3: 更容易出卡通光影，构图和花纹装饰性强

• loha04_old3: 构图容易出全身，光影保留效果好

• loha06_old1: 对构图限制小，人物动作自然

• loha02_old2: 背景层次更多，对称性强

Dataset labeling and loha training method, see wiki link

https://aigc.ioclab.com/sd-showcase/chinese-ornament.html

Chinese Ornament Wiki in English

::: This page is an introduction to the use of traditional Chinese pattern Lora, for detailed training processes and layer debugging methods, please refer to our wiki. https://aigc.ioclab.com/sd-showcase/chinese-ornament.html :::

Chinese traditional pattern Lora by IoC Lab

Training and Data: @shichen @ciaochaos

Version Description 2023.4.1 Uploaded Lora, Introduction Content (This Article) 2023.4.4 Updated Training Method (Wiki To Be Uploaded)

In the pictures, you can see more than 25 versions. I selected the most distinctive versions for you to download, or you can go to Huggingface to download all versions.

ioclab/IocLab_Models at main (huggingface.co)

How to use:

In the annotation of the training set, the following five language segments are divided:

traditional chinese pattern

with circular symmetrical composition
with square symmetrical composition
with special-shaped symmetrical composition

red and blue color scheme

featuring horses and clouds
featuring flowers and branches

a flower in the center

So you can expand around these five fixed sentence patterns. If you need to use it as a background, you can reduce the weight and add one girl.

Layered Decomposition:

As shown in the layered images, under high weight, by only turning on the weights of midd, outd, and other Lora layers, you can achieve a person-background image with a style more retained than directly reducing the weight.

Recommended Model Version Introduction:

Here are some model versions with distinctive effects, but the output effect is related to parameters, seeds, etc. You can try them out and we look forward to your comments.

Distinctive Model Versions

Here are some distinctive model versions:

• 4: The characteristic is that the pattern is more cartoonish and the color is more gentle.

• 7: The characteristic is that it better restores the keywords and is more in line with the tendency of the dataset.

• 10: The characteristic is that it is more symmetrical, and the composition tends to be non-centrally symmetrical.

• 13: The characteristic is that compared to 10, the color will be more gentle, and the third scattered composition will appear under high weight.

• 15: The characteristic is high saturation and high horizontal symmetry composition.

• 16: The characteristic is high saturation + scattered composition.

Character Matching Versions

The following are versions that are easy to match with characters:

• loha01_old3: More likely to produce cartoon light and shadow, with strong composition and decorative patterns.

• loha04_old3: The composition is easy to produce the whole body, and the effect of retaining light and shadow is good.

• loha06_old1: The restriction on composition is small, and the character's movements are natural.

• loha02_old2: More layers in the background, strong symmetry.

For dataset annotation and Lora training methods, please refer to the wiki link.

https://aigc.ioclab.com/sd-showcase/chinese-ornament.html

Weight : 0.4～0.6?

The model for the illustration for the sample is based on 7th_anime_v3_A and has been merged several times, including AOM3.

7th_anime_v3_A will be a close illustration.

Please use it in an environment where LoCon is available.

AnyLoRA is used as the model for the study.

A Character LoCon for Ijichi Seika (伊地知星歌) | Bocchi the Rock!

She may not be as popular as other characters in Bocchi the Rock, but I still like her.

• Recommeneded prompts: red eyes,long hair,ahoge,blonde hair,black choker,off shoulder,bangs (Check my sample pics for detail)

• Trigger word: red eyes, blonde hair

• Recommeneded Weight: 0.6-1 (You can adjust it by your perference.)

• CFG scale: 5.5-7

• Check point recommendation: Anything V5, DanMix

• How to use LoCon/LyCORIS? Please read instruction of a1111-sd-webui-locon.

• This LoCon still have some small issues, I may fix it later.

If you like my work,please give me a like <3.

If you like my model please give me a like <3.

If you like my model, please click like <3.

「May you still remember the moonlight……」

• This is a lycoris model, it requires the locon extensions to be installed.

• The trigger words for alternative costumes are still under active testing and may not be stable

• Recommened prompts: 1girl, solo, chain necklace\(arcueid\), black pantyhose, white long sleeves turtleneck sweater, blue pleated skirt, looking at viewer, brown footwear, knees boots

• Weight 0.7-0.9 works best for me.

• CFG Scale: 5.5-7

Model based on Erkerut art style. Works best with MeinaHentai (v2 used for example images) at 0.9.

Personally, I feel that Mr. Babaobei’s painting style is very different from the early and late stages. I chose a late-stage painting style.

I use anythingv4.5, weight 0.5-0.9. Better to use Hires. Fix.

Tigger words use Happoubi Jin, which can trigger the painting style.

I made a few main characters by the way, you can use Happoubi Jin + character name to summon them.

The characters that can be summoned are: Tsukioka Karen, Dorothy, Alice, Alice, Kagarino Kirie, Inamori Haru, and Sawatari Shizuku. Among them, Dorothy needs to be used in combination with short hair, and other names can be used. When summoning a character, the recommended weight is 0.7-0.9.

I feel that there is a significant difference in the before and after style of Happoubi Jin's artwork, and I chose the processing style.

I used anythingv4.5, with a weight of 0.5-0.9. It is recommended to use Hires. Fix.

Trigger words: Happoubi Jin, which can trigger the art style.

I also made several main characters, and you can summon them using Happoubi Jin + character name. The characters that can be summoned are: Tsukioka Karen, Dorothy, Alice, Kagarino Kirie, Inamori Haru, and Sawatari Shizuku. Dorothy needs to be used with "short hair" to summon, while the others can be summoned using their names. It is recommended to use a weight of 0.7-0.9 when summoning characters.

Hitoimim Style

An Artstyle LoHa

V2b is out!

Important: I only recommend using this LoHa to generate upper body images for the best results. (I only recommend using this LoHa to generate upper body images for the best results.)

• Recommeneded prompts: simple background,upper body,cropped shoulders,hair ornament,ribbon,bow,jewelry (Please check my sample work for more details)

• Trigger word: you don't need any trigger word for this LoHa

• Recommeneded Weight: 0.6-1 (You can adjust it by your perference.)

• CFG scale: 5.5-7

• Check point recommendation: Anything V4.5, Dan mix

• This LoHa may have some small issue, I will fix it later!

• How to use LoHa/LyCORIS? Please read instruction of a1111-sd-webui-locon.

If you like my work,please give me a like <3.

If you like my model please give me a like <3.

If you like my model, please click like <3.

When you don't just want bigger; you want more too. Example images generated from 1 set of 8 using AnythingV4.5

300+ Ayaka HQ ​​images dataset

A Character LoCon for mao isara [Ensemble Stars!]

Weight:0.4～0.7

Hires. fix

trigger words: isara mao\(ensemble stars!\)

*Please excuse me for the following in Japanese.

運要素が絡みますがPoi衣装を出現させることは確認しています。

Trickstar Costume (! Era)

Appearance Up Spells: plaid_red jacket, white shirt, red_plaid pants, black gloves, purple wristband, jewelry,

Trickstar Outfit (!! Era)

Appearance increase spells: plaid_red jacket, white shirt, black pants, black gloves, fingerless gloves, jewelry,

Uniform (winter)

Appearance rate up spells: hood, necktie, white shirt, open blazer, blue blazer, school uniform, plaid, pants,

Haircut (usually hair ornaments, hairclip, bangs pinned back, even if you don't specify the spell, the hairpin and bangs will be put up, so if you want to do it)

Appearance rate up spell: down hair, hair between eyes

Negatively, hair ornaments, hairclips, and bangs pinned back can be used to make the hair look a little lower.

I can't say that it's easy to use, but I'd appreciate it if you could use it.

weight: 0.6-0.8

recommend prompt: pauldrons