Stable Diffusion Xl

Pony: People's Works V1 V6 V2 Ani3.1

model_id: pony-people-s-works-v1-v6-v2-ani3-1

<p>Base model: Animagine v3.1</p><p>batch size=12</p><p>dimension=32 conv_dim=8</p><p>alpha=16 conv_alpha=2</p><p>learning rate: U-net lr=6e-4 Te lr=2e-4</p><p>optimizer: AdamW</p><p><strong>训练策略：</strong></p><p>训练数据集：374张Cvitai上的pony v6图片，393张正则图片</p><p>训练总共21个epoch，每个epoch中重复5次训练图片、2次正则图片</p><p>使用了 <em>cosine with restarts </em>调度器，第一个epoch（323步）作为预热；剩余的20个epoch分为5个循环，每个循环中有4个epoch（也就是说每个cycle里训练了20次训练图片）</p><p>每张图片训练了5+5×4×5=105次。</p><p><strong>Training strategy: </strong></p><p>Training dataset: 374 selected pony v6 pics from Civitai, 393 regularization images</p><p>Trained 21 epochs, each epoch repeated training training data 5 times and reg data 2 times.</p><p>Applied <em>cosine with restarts</em> scheduler, first 1 epoch (323 steps) for warmup and divided rest 20 epochs into 5 cycles, each cycle contained 4 epochs (training images were trained 20 times per cycle).</p><p>Trained 5+5×4×5=105 times for one pic totally.</p>