这是一个经过500万张图像训练而成的XL大模型，内置了超过2000个强风格标签。 This is an XL model trained on 5 million images, featuring over 2,000 strong style tags.模型的训练过程包括了多阶段训练： The training process includes multiple stages:<ul><li>使用A40双卡训练了400小时</li><li>Trained for 400 hours with dual A40 GPUs</li><li>通过4090训练了1104小时来完善底模</li><li>Enhanced with 1104 hours of training using the 4090 GPU</li><li>最后在双卡A100 80G上进行了800小时的画风训练</li><li>Finally, completed 504 hours of style training on dual A100 80G GPUs</li></ul>目前正式版1.0仍有部分标签欠拟合，后续版本会进行修复。 Version 1.0 still has some underfitted tags, which will be fixed in future updates.特别感谢rnglg2在算力和数据处理方面的支持，以及且听风吟、Willy、青秋、nano、Cyanelis Deuxieme、群阿没、复读机bot在训练集构建方面的帮助。 Special thanks to rnglg2 for computational and data processing support, and to 且听风吟, Willy, 青秋, nano, 群阿没、Cyanelis Deuxieme, and 复读机bot for assistance in building the training dataset.训练底模来自<a target="_blank" rel="ugc" href="https://civarchive.com/models/264323">【SDXL】/Anime/bulldozer_BETA - v2.0 | Stable Diffusion XL Checkpoint | Civitai </a><hr /><h3 id="(usage-recommendation)-tp0bkwty2">模型的使用建议 (Usage Recommendation)</h3>推荐CFG不超过5我们对数据集进行了美学评分，评分标准如下： We applied aesthetic scoring to the dataset, with the following rating criteria:<ul><li>Core &gt; 0.75: 质量标签 = "masterpiece"</li><li>Core &gt; 0.75: Quality Tag = "masterpiece"</li><li>0.6 &lt; score &lt;= 0.75: 质量标签 = "high quality"</li><li>0.6 &lt; score &lt;= 0.75: Quality Tag = "high quality"</li><li>0.5 &lt; score &lt;= 0.6: 质量标签 = "normal quality"</li><li>0.5 &lt; score &lt;= 0.6: Quality Tag = "normal quality"</li><li>0.3 &lt; score &lt;= 0.5: 质量标签 = "low quality"</li><li>0.3 &lt; score &lt;= 0.5: Quality Tag = "low quality"</li><li>score &lt;= 0.3: 质量标签 = "worst quality"</li><li>Score &lt;= 0.3: Quality Tag = "worst quality"</li></ul>正常使用时，只需在标签前添加<code>masterpiece</code>, <code>best quality</code>, 或<code>high quality</code>即可。 For normal use, simply add <code>masterpiece</code>, <code>best quality</code>, or <code>high quality</code> before the tag.风格标签的完整表格将在完整版发布时提供，目前可以通过参考示例图来使用。 The full table of style tags will be provided in the full release. For now, you can refer to the example images.<a target="_blank" rel="ugc" href="https://rnglg2-my.sharepoint.com/:u:/g/personal/hans_rnglg2_onmicrosoft_com/EdTDVlmCZaZKlsoi_AOQjOsBW4xPx6H7S1XJWh8jYNH3aw?e=xmbUFF">https://rnglg2-my.sharepoint.com/:u:/g/personal/hans_rnglg2_onmicrosoft_com/EdTDVlmCZaZKlsoi_AOQjOsBW4xPx6H7S1XJWh8jYNH3aw?e=xmbUFF</a>训练参数 (Training Parameters)resolution = "1024,1024"enable_bucket = truemin_bucket_reso = 256max_bucket_reso = 1536bucket_reso_steps = 32output_dir = "/root/"save_model_as = "safetensors"save_precision = "fp16"save_every_n_epochs = 2max_train_epochs = 20train_batch_size = 5gradient_checkpointing = falselearning_rate = 0.00003learning_rate_te1 = 0.000001learning_rate_te2 = 0.000001lr_scheduler = "cosine_with_restarts"lr_scheduler_num_cycles = 20optimizer_type = "AdamW"min_snr_gamma = 5sample_every_n_epochs = 1log_with = "tensorboard"logging_dir = "./logs"caption_extension = ".txt"shuffle_caption = trueweighted_captions = falsekeep_tokens = 4max_token_length = 255multires_noise_iterations = 8multires_noise_discount = 0.4no_token_padding = falsemixed_precision = "bf16"full_bf16 = truexformers = truelowram = falsecache_latents = truecache_latents_to_disk = truepersistent_data_loader_workers = truetrain_text_encoder = true<h3 id="(disclaimer)-sa8dbc4mt">免责声明 (Disclaimer)</h3>请遵守当地法律法规，以免造成麻烦。鉴于模型的实际用途不受模型作者控制，因模型输出的图片所产生的一切后果由图片输出者自行承担。基于此模型训练出的衍生模型请标明出处。 As the actual use of the model is beyond the control of the model creators, all consequences arising from images generated by this model are the sole responsibility of the user.<h3 id="-84jtuvo4i"></h3>

Files with SHA256: eaf4363792e17af8b81d3265866de038b3b530d4e261a7469dc8ba9dfa496258