A game artwork style model like Blue Archive that based on <a target="_blank" rel="ugc" href="https://huggingface.co/cagliostrolab/animagine-xl-3.1">Animagine-XL-3.1</a><h2 id="heading-185">Prompt Guidelines/プロンプト記法</h2>Almost same as Animagine-XL-3.1( <a target="_blank" rel="ugc" href="https://cagliostrolab.net/posts/animagine-xl-v31-release">https://cagliostrolab.net/posts/animagine-xl-v31-release</a> )Animagine-XL-3.1( <a target="_blank" rel="ugc" href="https://cagliostrolab.net/posts/animagine-xl-v31-release">https://cagliostrolab.net/posts/animagine-xl-v31-release</a> )と同じです。<h2 id="heading-186">Note/注意</h2>Halos may be generated unintentionally due to the dataset.データセットの影響で意図せずヘイローが生成される場合があります。A separate VAE is not required.別途VAEを使用する必要はありません。<h2 id="heading-187">Training information:</h2>Finetuned Anmagine-XL-3.1 by repeating the training and merging a LoRA two times with sd-scripts.sd-scriptsで二回loraの学習とマージを繰り返すことでファインチューンされました。Basic training parameters:Resolution:1024x1024Bucketing: enabled,min512,max2048LR scheduler: cosine_with_restartsOptimizer: LionKeep tokens:0Network dim/alpha: 16/6Conv dim/alpha:8/2Gradient checkpointing:trueGrad acc steps:1Batch size: 21st LoRA parameters:Epochs:6Unet learning rate:0.0001TE learning rate:8e-06LR scheduler cycles:5Shuffle caption:trueSeed:42LR warmup steps:2502nd LoRA parameters:Epochs:16Unet learning rate:5e-05TE learning rate:4e-06LR scheduler cycles:9Seed:43LR warmup steps:250<h2 id="heading-188">Dataset information:</h2>Dataset size: 333Repeats: 2<h2 id="heading-189">Training script(source code)</h2>sd-scripts( <a target="_blank" rel="ugc" href="https://github.com/kohya-ss/sd-scripts">https://github.com/kohya-ss/sd-scripts</a> )<h2 id="heading-190">Notice/おしらせ:</h2>This model is licensed under the Fair AI Public License 1.0-SD( <a target="_blank" rel="ugc" href="https://freedevproject.org/faipl-1.0-sd/">https://freedevproject.org/faipl-1.0-sd/</a> ).If you make modify this model, you must share both your changes and the original license.このモデルはFair AI Public License 1.0-SD( <a target="_blank" rel="ugc" href="https://freedevproject.org/faipl-1.0-sd/">https://freedevproject.org/faipl-1.0-sd/</a> )でライセンスされています。このモデルを改変する場合、変更点および元のライセンスを明記する必要があります。

Turquoise XL

Trained with DoRA + Scheduled Huber Loss

Files with SHA256: 1cb3ef9f582f2b2fb45b368ced6f6b6fd362ee160888c08d58c8724ba5567150