<code>Useless Detail Enhancer</code> 是一个训练失败的 Stable Diffusion XL LoRA 模型。它通过调整模型权重来添加大量无效的细节，可能会让画面显得更加混乱。使用触发词“Best-A”可以激活这一效果。适合那些想要探索模型极限或尝试一些有趣效果的用户。请谨慎使用，因为它可能会对画面产生不可预期的影响。 基于T-PonyNai3 5.5训练 训练参数为 - <code>train_batch_size = 55</code>- <code>gradient_checkpointing = true</code>- <code>gradient_accumulation_steps = 10</code>- <code>unet_lr = 0.001</code>- <code>text_encoder_lr = 0.0001</code>- <code>lr_scheduler = "cosine_with_restarts"</code>- <code>lr_warmup_steps = 0</code>- <code>lr_scheduler_num_cycles = 1</code>- <code>optimizer_type = "Lion8bit"</code>- <code>multires_noise_iterations = 20</code>- <code>multires_noise_discount = 0.75</code> 训练参数仅供参考-不要尝试使用我的参数，你会百分百训练失败！ ---Useless Detail Enhancer is a failed training Stable Diffusion XL LoRA model. It adjusts model weights to add a large amount of useless details, which may make the image appear more chaotic. Use the trigger word “Best-A” to activate this effect. It is suitable for those who want to explore the limits of the model or try some interesting effects. Use with caution, as it may produce unpredictable impacts on the image.Based on T-PonyNai3 5.5 trainingTraining parameters:- <code>train_batch_size = 55</code>- <code>gradient_checkpointing = true</code>- <code>gradient_accumulation_steps = 10</code>- <code>unet_lr = 0.001</code>- <code>text_encoder_lr = 0.0001</code>- <code>lr_scheduler = "cosine_with_restarts"</code>- <code>lr_warmup_steps = 0</code>- <code>lr_scheduler_num_cycles = 1</code>- <code>optimizer_type = "Lion8bit"</code>- <code>multires_noise_iterations = 20</code>- <code>multires_noise_discount = 0.75</code>Training parameters are for reference only—do not attempt to use my parameters, you will definitely fail!

Files with SHA256: 9183dbe030d468e36dd3028dced2e298cb3e88f618aca279bd0a28850f6a61d1