<h3 id="about-this-model-(v2.1)-4gq6nv1g3">### About this model (v2.1)</h3><ul><li>Base merge:<ul><li>Illustrious-XL-v2.0-stable</li><li>glora trained on ~260 images 2048px (v1.1s outputs)</li></ul></li><li>Enhancement:<ul><li>Slight foot-balance correction at 1536-2048px</li></ul></li><li>Training:<ul><li>LyCORIS=glora, Batch_size=2, epochs=50(6500steps), DIM=24, Alpha=24 Unet_lr=2e-5, Text_encoder_lr=1e-06, Optimize=CAME, scheduler=cosine_with_min_lr=0.2, warmup=0.05, max_grad_norm=1, Scale_weight_norms=1</li></ul></li><li>VAE:<ul><li>baked-in illustrious-xl-v2.0.safetensors</li></ul></li></ul><h2 id="v2.1-vrh7wbwvs">v2.1</h2>高解像度での足のバランスを少し修正(2048)Slightly corrected foot balance in high resolution(2048)<h2 id="v2.0-h2550k48o">v2.0</h2>(en)This model is a merger of glora trained on approximately 220 high-resolution (2048) images generated by v1.0 and v1.1 of this model, merged with Illustrious-XL-v2.0-stable. As this is primarily a test of how well images hold up at resolutions of 1024 and above, there are no particular characteristics of the model to note.The showcase images have not been processed with Hires.fix or similar methods.Example resolutions tested:<ul><li>2048x1536</li><li>1344x1728</li><li>2048x2048</li><li>1824x1248</li><li>1536x1536</li></ul>⚠Perfect results are not guaranteed at these resolutions.(ja)このモデルのv1.0とv.1.1で生成した2048解像度の画像、約220枚で学習したgloraをIllustrious-XL-v2.0-stableにマージしたものです。1024以上の解像度でどれだけ崩れないかのテスト的な側面が強いので、モデルの特徴などは特に有りません。ショーケースの画像はHires.fix等の処理はしていません。テストした解像度例:<ul><li>2048x1536</li><li>1344x1728</li><li>2048x2048</li><li>1824x1248</li><li>1536x1536</li></ul>⚠これらの解像度で常に完全な結果が得られるとは限りません。<hr /><h2 id="v1.0-v1.1s-ikornungi">v1.0-v1.1s</h2>This is a model fixed to my preferred style. The colors are lighter, and I adjusted the way the skin reflects light to my liking.Even at 1536x1536 resolution (or 1344x1728, etc.), it breaks down less than Illustrious-XL-v0.1, but depending on the composition and pose, the body proportions can become awkward.With full-body compositions, the eyes become unstable.Baked <a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/sdxl-vae/blob/main/sdxl_vae.safetensors">sdxl_vae.safetensors</a>The "s" in v1.1s stands for simple and stable.<h2 id="merge-models(v1.0):-8hanery4c">Merge Models(v1.0):</h2><h2 id="1.-wai-nsfw-illustrious-sdxl(v13.0)-bxn5ltzki">1. <a target="_blank" rel="ugc" href="https://civarchive.com/models/827184/wai-nsfw-illustrious-sdxl?modelVersionId=1612720">WAI-NSFW-illustrious-SDXL(v13.0)</a></h2><h2 id="2.-paruparu_illustrious(v3.2)-x5w2uhs57">2. <a target="_blank" rel="ugc" href="https://huggingface.co/paruparu95483/paruparu_illustrious">paruparu_illustrious(v3.2)</a></h2><h2 id="3.-diving-illustrious-anime(v7.0)-qvvdkcv8d">3. <a target="_blank" rel="ugc" href="https://civarchive.com/models/1170176/diving-illustrious-anime?modelVersionId=1611024">Diving-Illustrious Anime(v7.0)</a></h2><h2 id="4.-nova-orange-xl(v8.0)-lbw074dgx">4. <a target="_blank" rel="ugc" href="https://civarchive.com/models/967405/nova-orange-xl?modelVersionId=1610030">Nova Orange XL(v8.0)</a></h2><h3 id="5.-2lora:1536x1536-resolution-training-vypssllcw">5. 2LoRA:1536x1536 Resolution training</h3><h3 id="thanks-to-all-the-model-creator!-6tx9xeqkz">Thanks to all the Model Creator!</h3><h3 id="-fwn6z6wqk"></h3><h3 id="commercial-use-is-prohibited-(based-on-diving-illustrious)-0p6rgztne">Commercial use is prohibited (based on Diving-Illustrious)</h3><h3 id="personal-setting:-byusf37v8">Personal setting:</h3><pre><code>masterpiece, best quality, very aesthetic, absurdres, </code></pre><pre><code>lowres, bad quality, worst quality, normal quality, very displeasing, bad anatomy, sketch, jpeg artifacts, unfinished, signature, watermark, text, ugly, </code></pre>Sampler:DPM++ 3M SDE CFG++Scheduler:SimpleCFG:1Steps:20-30<hr />自分好みのスタイルに固定したマージモデルです。色合いは薄めで、肌の光り方を好みに調整しました。1536x1536解像度(1344x1792など)でもIllustrious-XL-v0.1よりは破綻しませんが、構図やポーズによって人体のバランスが変になります。全身の構図だと目が安定しません。<a target="_blank" rel="ugc" href="https://huggingface.co/stabilityai/sdxl-vae/blob/main/sdxl_vae.safetensors">sdxl_vae.safetensors</a>を焼き込んであります。v1.1sの"s"はシンプルと安定の略です。<h2 id="(v1.0):-yozft1i5l">マージモデル(v1.0):</h2><h2 id="1.-wai-nsfw-illustrious-sdxl(v13.0)-20rmpa3re">1. <a target="_blank" rel="ugc" href="https://civarchive.com/models/827184/wai-nsfw-illustrious-sdxl?modelVersionId=1612720">WAI-NSFW-illustrious-SDXL(v13.0)</a></h2><h2 id="2.-paruparu_illustrious(v3.2)-v0kiy87ez">2. <a target="_blank" rel="ugc" href="https://huggingface.co/paruparu95483/paruparu_illustrious">paruparu_illustrious(v3.2)</a></h2><h2 id="3.-diving-illustrious-anime(v7.0)-fq6gy5i16">3. <a target="_blank" rel="ugc" href="https://civarchive.com/models/1170176/diving-illustrious-anime?modelVersionId=1611024">Diving-Illustrious Anime(v7.0)</a></h2><h2 id="4.-nova-orange-xl(v8.0)-jigj8zm36">4. <a target="_blank" rel="ugc" href="https://civarchive.com/models/967405/nova-orange-xl?modelVersionId=1610030">Nova Orange XL(v8.0)</a></h2><h3 id="5.-2lora:1536x1536-kn5adetyr">5. 2LoRA:1536x1536解像度で学習</h3><h3 id="-8bglcvzqv">全てのモデル制作者に感謝します！</h3><h3 id="-pynj0u6ps"></h3><h3 id="(diving-illustrious)-dgpo39joa">商用利用禁止です (Diving-Illustriousに基づく)</h3><h3 id=":-rc45c6ffb">個人的な設定:</h3><pre><code>masterpiece, best quality, very aesthetic, absurdres, </code></pre><pre><code>lowres, bad quality, worst quality, normal quality, very displeasing, bad anatomy, sketch, jpeg artifacts, unfinished, signature, watermark, text, ugly, </code></pre>Sampler:DPM++ 3M SDE CFG++Scheduler:SimpleCFG:1Steps:20-30

AlusterMix iL

<ul><li>glora trained with 230 images of 2048 resolution generated by v1.0 and v1.1s lora merged into illustriousXLV20_v20Stable</li><li>Stability itself is higher in the previous version.</li><li>merged v0.1 and v1.1 based models in illustriousXLV20_v20 will collapse at high resolution, so we adopted a distilled approach</li><li>v1.0とv1.1sを使用して2048解像度生成した画像230枚を学習したgloraをillustriousXLV20_v20Stableにマージしたものです</li><li>1536から2048解像度のテスト的な側面が強いので、安定性自体はv1系統の方が高いです</li><li>illustriousXLV20_v20にv0.1とv1.1ベースのモデルを混ぜると高解像度では崩れるので、蒸留的な手法を採用しました</li></ul>

Files with SHA256: d1cd5bd1ca9fd4f31f1f8eef9212cc1f1a3669bafe35dd8833a0312aa3ef84d3