(日本語説明は後半にあります)v7.0 Update : Made detailed adjustments and modifications at each block(IN, OUT, MID) and text encoder to improve background quality, composition, and overall finish.<h2 id="about-this-model">About this model</h2>"IllusionBreed" is an Illustrious-based SDXL checkpoint, pursuing "kawaii", cuteness.There are 2 variations for v7.0 :<ul><li>Basic 2D painting style</li><li>"Slender" : 2D painting and more slender body style</li><li>To download each types, switch version tabs ↑ above.</li></ul>I feel that model merging is similar to the breeding of flowers (to improve their appearance), so named "Breed".These models have special features :<ul><li>Include many originally trained data to adjust painting style (saturation, tone), eye shape, and quality (finely detailed pupils, finely detailed outlines, fine painting, reducing grain noise etc).</li><li>Detailed backgrounds.</li><li>Powerful prompting based by Illustrious (and NoobXL).</li></ul>More various other LoRA's and Checkpoints are released on my <a target="_blank" rel="ugc" href="https://civarchive.com/user/JujoHotaru/models">CivitAI models list</a> or <a target="_blank" rel="ugc" href="https://huggingface.co/JujoHotaru/lora">HuggingFace repository</a>. Please check it out.<hr /><h2 id="recommended-output-settings">Recommended output settings</h2><ul><li>CFG Scale : 4-7 (recommended : 5)</li><li>Sampler : Eular A</li><li>Sampling steps : 20 - 60 (recommended : 40)</li><li>Resolution : mainly tested on 768x768, 768x1152, 832x1216, 800x1200 and 1024x768</li><li>Using A-Detailer (for face etc.) is highly recommended. (inpaint denoising strength : 0.3 for face, 0.5 for breasts, etc.)</li><li>HiRes Fix : use SwinIR_4x, R-ESRGAN 4x+ or 4x-UltraSharp with denoise strength 0.3-0.7.</li><li>Quality prompt / Negative prompt : See below.</li><li>VAE : embedded.</li></ul><hr /><h2 id="quality-prompts-sample">Quality prompts sample</h2>This and base checkpoints are derived from Illustrious, so use quality prompts for Illustrious models.<pre><code>masterpiece, best quality, amazing quality, very aesthetic, absurdres, perfect anatomy,
perfect symmetrical face, cute face, perfect hands, detailed eyes, detailed hair, slender,
(PUT YOUR PROMPTS HERE)</code></pre>To improve background quality, <code>detailed background, photo background, blurry background</code> prompt is recommended.<h2 id="negative-prompts-sample">Negative prompts sample</h2><pre><code>lowres, bad quality, worst quality, bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn, blurry, watermark,
bad eyes, bright pupils, white pupils, bad hands, bad face,
(PUT YOUR NEGATIVE PROMPTS HERE)</code></pre><hr /><h2>このデータモデルについて</h2>v7.0 アップデート：IN/OUTの各階層ごとに細かく変更・調整を加え、背景品質や構図・仕上がりなどを改善しています。IllusionBreedは、二次元イラストレーション生成に特化した「Illustrious」派生形モデルで、可愛さを追求しています。v7.0では、以下2種類のバリエーションが用意されています：<ul><li>基本2Dタイプ</li><li>Slender : 2Dタイプ、若干頭身が高めのボディースタイル</li><li>各タイプをダウンロードするには、ページ上部のバージョンタブを切り替えてください。</li></ul>私はモデルのマージはお花の交配（品種改良）と近いものを感じているため、作成したデータモデルの名前には品種改良を意味する「Breed」を付けています。 Breedシリーズの特徴として、作者が独自に学習させた多数のLoRAをマージすることで、画質・色味・陰影や眼の形などを調整したり、輪郭線および眼の質感向上や粒子ノイズ低減、HiRes Fixの高品質化などを図っています。<h3>使い方</h3><ul><li>推奨される出力設定や品質指定プロンプト・ネガティブプロンプトなどは上記を参照してください。</li><li>背景の描写品質を上げたい場合、背景関連のプロンプトに <code>detailed background, photo background, blurry background</code> を追加することを推奨します。</li><li>サンプラーに「Eular A」系統以外を使うと画質が劣化しやすいので注意してください。</li><li>ネガティブプロンプトは空にせず、低品質系のワードを入れて調整したほうが良い画質が出やすいようです。（高速出力などのため使えない場合は無くても大丈夫です）</li><li>A-Detailerの使用を推奨します。顔のリファインのInpaint denoising strengthは0.3、裸体の胸には0.5程度がよいでしょう。</li><li>ネコミミなど動物耳キャラクターを生成する場合、人間耳を出さないようにする<a target="_blank" rel="ugc" href="https://civarchive.com/models/1505391/human-ear-eraser-for-animal-ear">人耳消去LoRA</a>も用意しています。</li></ul>VAEは焼き込み済みですので、別途用意する必要はありません。他にもさまざまなLoRAやデータモデルを公開していますので、<a target="_blank" rel="ugc" href="https://civarchive.com/user/JujoHotaru/models">CivitAIの配布モデルリスト</a>または<a target="_blank" rel="ugc" href="https://huggingface.co/JujoHotaru">HuggingFaceリポジトリ</a>をご覧ください。<hr /><h2 id="merge-recipe-(for-v1.0)">Merge recipe (for v1.0)</h2><ul><li>Based on 8 models (all merged with ratio 0.125) :<ul><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/894218">copycatNoob_v10</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/869526">illuminaXLTypemModified_11</a></li><li>illuminous_v01</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/880866">illustPossible_v10</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/827460">librasillustriousxl_v40</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/820208">obsessionIllustrious_v30</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/853496">simpleIllustrious_valt</a></li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/910006">zukiILL_v10</a></li></ul></li><li>Original finetuned data with <a target="_blank" rel="ugc" href="https://civarchive.com/models/783194">VioletBreed</a> / <a target="_blank" rel="ugc" href="https://civarchive.com/models/814740">RoseBreed</a> outputs</li><li>Original trained LoRAs for eye shape/painting/detailing</li><li><a target="_blank" rel="ugc" href="https://civarchive.com/models/152040">XL_VAE_C F2</a> VAE</li></ul>

IllusionBreed

Updated text encoder (merged with ILL2.0) and tuned output quality.

illusionbreed_v60.safetensors

IllusionBreed v6.0

v6.0

"Slender" v7.0

v7.0

"2.2D Slender" v6.0

"Slender" v6.0

"2.2D" v6.0

"2.2D Slender" v5.0

"2.2D" v5.0

"Slender" v5.0

v5.0

"2.2D Slender" v3.0

"Slender" v3.0

"2.2D" v3.0

v3.0

"Cute" v2.0

"2.2D" v2.0

v2.0

"2.5D" v1.0

v1.0

Created by JujoHotaru"IllusionBreed" is an Illustrious-based SDXL checkpoint, pursuing "kawaii", cuteness.v1.0 is an experimental version.This checkpoint has 2 variations : <a target="_blank" rel="ugc" href="https://civitai.com/models/935880?modelVersionId=1047621">2D style (basic) type</a> or <a target="_blank" rel="ugc" href="https://civitai.com/models/935880?modelVersionId=1047628">a bit realistic (2.5D) style type</a>. To download each types, switch version tabs ↑ above.I feel that model merging is similar to the breeding of flowers (to improve their appearance), so named "Breed".These models have special features :<ul><li>Include many originally trained data to adjust painting style (saturation, tone), eye shape, and quality (finely detailed pupils, finely detailed outlines, fine painting, reducing grain noise etc).</li><li>Detailed backgrounds.</li><li>Powerful prompting based by Illustrious (and NoobXL).</li></ul>More various other LoRA's and Checkpoints are released on my <a target="_blank" rel="ugc" href="https://civitai.com/user/JujoHotaru/models">CivitAI models list</a> or <a target="_blank" rel="ugc" href="https://huggingface.co/JujoHotaru/lora">HuggingFace repository</a>. Please check it out.Recommended output settings<ul><li>CFG Scale : 4-5</li><li>Sampler : Eular A</li><li>Sampling steps : 20 - 40</li><li>Resolution : mainly tested on 768x768, 768x1152, 832x1216, 800x1200 and 1024x768</li><li>Using A-Detailer (for face etc.) is highly recommended. (inpaint denoising strength : 0.3-0.5)</li><li>HiRes Fix : use SwinIR_4x, R-ESRGAN 4x+ or 4x-UltraSharp with denoise strength 0.3-0.7.</li><li>Quality prompt / Negative prompt : See below.</li><li>VAE : embedded.</li></ul>Quality prompts sampleThis and base checkpoints are derived from Illustrious, so use quality prompts for Illustrious models.masterpiece, best quality, amazing quality, very aesthetic, absurdres, cute, kawaii, cute round face, slender, detailed eyes, detailed hair, (PUT YOUR PROMPTS HERE)Negative prompts samplelowres, bad quality, worst quality, bad anatomy, sketch, jpeg artifacts, ugly, poorly drawn, blurry, watermark, bad eyes, bright pupils, white pupils, ugly, bad hands, bad face, (PUT YOUR NEGATIVE PROMPTS HERE)

Files with SHA256: 08a1b4c1e69704da6919f1dab141357b23f13fdf0edfedabfa8f0a53593b95b3