<h2>もうフィギュア人形を洗う必要はありません！</h2><h3>■概要</h3>フィギュア人形にぶっかけるLoraです（Wan2.2 I2V 14B）。フィギュア人形以外の画像に対してもExcessive Bukkake Loraとして使えます。プロンプト例：The penis repeatedly ejaculates into the figure doll, which is completely fixed and motionless. Her face and whole body are covered with a large amount of translucent white semen, and the semen drips slowly.<h3>■使用モデル</h3>元画像にはIllustrious用の<a target="_blank" rel="ugc" href="https://civarchive.com/models/1341732">cum_on_figure Lora</a>を使っています。サンプル動画の多くは<a target="_blank" rel="ugc" href="https://civarchive.com/models/1981116/dasiwa-wan-22-i2v-14b-or-lightspeed-or-safetensors">DaSiWa WAN 2.2 v10</a>で生成しています。他のモデルだと人形が動きやすいです。<a target="_blank" rel="ugc" href="https://civarchive.com/models/1946997/wan-22-male-finishing-anime?modelVersionId=2579567">male finishing Lora</a>との併用も効果的です。<h3>■学習方法</h3>Loraの作成に当たって、<a target="_blank" rel="ugc" href="https://civarchive.com/articles/27705/wan22-i2v-and-ai-toolkit-beginners-training-guide-for-local-pc-training">"WAN2.2 i2v" and "AI-Toolkit" Beginner's Training Guide</a> を参考にしました。著者に感謝します。ただし、私のローカル環境ではスペック不足のためrunpod上でAI toolkitを動かして作成しました。16fpsの81frame動画素材を20個用意し、RTX PRO 6000で16時間、1750step学習しました。30ドルほどかかりましたが、新しいフィギュア人形を買うより安く済みました。<hr /><h2 id="&quot;you-don't-have-to-wash-figure-dolls-anymore.&quot;">"You don't have to wash figure dolls anymore."</h2><h3 id="overview">■ Overview</h3>This is a LoRA specialized for bukkake on anime scale figures (trained on Wan2.2 I2V 14B).It can also be used as a general Excessive Bukkake LoRA on subjects other than figure dolls.Prompt example:<pre><code>The penis repeatedly ejaculates onto the completely motionless anime figure doll. Her face and entire body are covered in thick translucent white semen, which slowly drips down.</code></pre><h3 id="recommended-model">■ Recommended Model</h3>You can make first image easily with the <a target="_blank" rel="ugc" href="https://civarchive.com/models/1341732">cum_on_figure Lora</a> made for Illustrious.Most sample videos were generated using <a target="_blank" rel="ugc" href="https://civarchive.com/models/1981116/dasiwa-wan-22-i2v-14b-or-lightspeed-or-safetensors">DaSiWa WAN 2.2 v10</a>. Other models tend to make the figure move more easily.Combining with <a target="_blank" rel="ugc" href="https://civarchive.com/models/1946997/wan-22-male-finishing-anime?modelVersionId=2579567">Male Finishing LoRA</a> is also very effective.<h3 id="training-details">■ Training Details</h3>This LoRA was created with reference to the <a target="_blank" rel="ugc" href="https://civarchive.com/articles/27705/wan22-i2v-and-ai-toolkit-beginners-training-guide-for-local-pc-training">"WAN2.2 i2v" and "AI-Toolkit" Beginner's Training Guide</a>. Big thanks to the authors.Due to insufficient specs on my local machine, training was done on RunPod using the official AI Toolkit.<ul><li>Dataset: 20 video clips (16fps, 81 frames each)</li><li>GPU: RTX PRO 6000</li><li>Steps: 1750 steps</li><li>Training time: Approximately 16 hours</li><li>Cost: About $30</li></ul>Still much cheaper than buying a new figure doll.

Files with SHA256: 64895de2931a0e1ed3e6955f663e2ee4d15f46f66d1d993b756013f5ca2d161f