FantasyParadise模型基于Animagine XL 3.1 训练而成延续了Animagine XL 3.1 出图方式和参数

推荐设置

要引导模型生成高美感图像，请使用负面提示，例如：

nsfw, lowres, (bad), text, error, fewer, extra, missing, worst quality, jpeg artifacts, low quality, watermark, unfinished, displeasing, oldest, early, chromatic aberration, signature, extra digits, artistic error, username, scan, [abstract]

为了获得更高质量的结果，请在提示前面加上：

masterpiece, best quality, very aesthetic, absurdres

建议使用大约 5-7 的CFG ，采样步骤低于 30，并使用 Euler Ancestral （Euler a） 作为采样器。

多方面分辨率

此模型支持生成以下维度的图像：

尺寸 纵横比

1024 x 1024	1：1 

1152 x 896	9:7

896 x 1152	7:9

1216 x 832	19:13

832 x 1216	13:19

1344 x 768	7：4 

768 x 1344	4：7 

1536 x 640	12：5 

640 x 1536	5：12

模型也完美继承了Animagine XL 3.1字符请参考https://huggingface.co/spaces/cagliostrolab/animagine-xl-3.1/blob/main/wildcard/character31.txt#L13

模型继承了Animagine XL 3.0，Animagine XL 3.1许可证

基于 Animagine XL 3.0，Animagine XL 3.1 属于 Fair AI 公共许可证 1.0-SD 许可证，该许可证与 Stable Diffusion 模型的许可证兼容。

要点：

修改共享：如果您修改 FantasyParadise_V1，您必须共享您的更改和原始许可证。

源代码可访问性：如果您修改后的版本可通过网络访问，请为其他人提供获取源代码的方法（如下载链接）。这也适用于派生模型。

分发条款：任何发行版都必须遵循此许可证或其他具有类似规则的许可证。

合规：不合规必须在 30 天内解决，以避免许可证终止，强调透明度和遵守开源价值观。

选择此许可证旨在保持 Animagine XL 3.1 的开放性和可修改性，与开源社区精神保持一致。它保护贡献者和用户，鼓励协作、合乎道德的开源社区。这确保了该模型不仅受益于公共输入，而且还尊重开源开发自由。

以下是训练参数

训练参数（Training parameters）

{

    "ss_sd_model_name": "animagine-xl-3.1.safetensors",

    "ss_resolution": "(1024, 1024)",

    "ss_clip_skip": "2",

    "ss_num_train_images": "6110",

    "ss_caption_dropout_every_n_epochs": "0",

    "sshs_legacy_hash": "f78d97b7",

    "ss_text_encoder_lr": "1e-05",

    "ss_multires_noise_iterations": "None",

    "ss_lowram": "False",

    "ss_zero_terminal_snr": "False",

    "ss_total_batch_size": "5",

    "ss_color_aug": "False",

    "ss_bucket_no_upscale": "False",

    "ss_ip_noise_gamma": "None",

    "ss_max_train_steps": "12220",

    "ss_num_batches_per_epoch": "1222",

    "ss_noise_offset": "None",

    "ss_scale_weight_norms": "None",

    "ss_keep_tokens": "0",

    "ss_num_epochs": "10",

    "ss_enable_bucket": "True",

    "ss_steps": "12220",

    "ss_optimizer": "bitsandbytes.optim.adamw.AdamW8bit",

    "ss_base_model_version": "sdxl_base_v1-0",

    "ss_lr_scheduler": "cosine_with_restarts",

    "ss_network_dropout": "None",

    "modelspec.prediction_type": "epsilon",

    "ss_learning_rate": "0.0001",

    "ss_network_alpha": "1.0",

    "ss_max_grad_norm": "1.0",

    "ss_min_bucket_reso": "256",

    "ss_gradient_checkpointing": "True",

    "ss_seed": "1026",

    "ss_face_crop_aug_range": "None",

    "ss_batch_size_per_device": "5",

    "ss_cache_latents": "True",

    "ss_shuffle_caption": "True",

    "modelspec.architecture": "stable-diffusion-xl-v1-base/lora",

    "modelspec.implementation": "https://github.com/Stability-AI/generative-models",

    "ss_sd_model_hash": "c798d390",

    "modelspec.sai_model_spec": "1.0.0",

    "ss_num_reg_images": "0",

    "ss_mixed_precision": "fp16",

    "ss_v2": "False",

    "ss_adaptive_noise_scale": "None",

    "ss_gradient_accumulation_steps": "1",

    "ss_caption_tag_dropout_rate": "0.1",

    "ss_output_name": "mk_kkk_xl",

    "sshs_model_hash": "0f6161a4645a9159bfdc2296d55f6c606bd37385d5b422d758220789d05e89ee",

    "ss_reg_dataset_dirs": {},

    "ss_max_bucket_reso": "4997",

    "modelspec.title": "mk_kkk_xl",

    "ss_flip_aug": "False",

    "ss_new_sd_model_hash": "e3c47aedb06418c6c331443cd89f2b3b3b34b7ed2102a3d4c4408a8d35aad6b0",

    "ss_full_fp16": "True",

    "ss_max_token_length": "225",

    "ss_training_started_at": "1711772016.6059477",

    "ss_min_snr_gamma": "None",

    "ss_random_crop": "False",

    "ss_training_finished_at": "1711798891.3724864",

    "ss_network_dim": "10000000",

    "ss_unet_lr": "0.0001",

    "ss_debiased_estimation": "True",

    "ss_lr_warmup_steps": "50",

    "ss_multires_noise_discount": "0.3",

    "ss_dataset_dirs": {

        "10_sc": {

            "n_repeats": 10,

            "img_count": 611

        }

    },

    "ss_sd_scripts_commit_hash": "77568ece5c8d4334cc204d34e148dd20f01c2348",

    "ss_epoch": "10",

    "ss_training_comment": "this LoRA model created from bdsqlsz by bdsqlsz'script",

    "modelspec.encoder_layer": "2",

    "modelspec.date": "2024-03-30T19:41:31",

    "ss_prior_loss_weight": "1.0",

    "ss_network_args": {

        "algo": "lokr",

        "use_scalar": "True",

        "train_norm": "True",

        "conv_dim": "4",

        "conv_alpha": "1",

        "use_tucker": "True",

        "preset": "attn-mlp",

        "factor": "8",

        "dropout": null

    },

    "ss_network_module": "lycoris.kohya",

    "modelspec.resolution": "1024x1024",

    "ss_session_id": "536501796",

    "ss_caption_dropout_rate": "0.0"

}

模型使用LyCORIS来模拟ΔW以用来模拟BASE 以下下是LyCORIS脚本命令

python merge.py AnimagineXL 31.safetensors mk_kkk_xl.safetensors FantasyParadise_V1.safetensors --is_sdxl --device cpu --dtype fp16 --weight 1.0

模型用途声明

1. 你不得将此模型及其衍生版本（如融合模型版本）托管于计划赚取收入或捐赠的网站/应用程序。如果您需要将此模型及其衍生版本用于商业目的（如生成式服务、售卖图片、将图片用于公开发表到短视频平台的文章或出版物等等）

2. 您不得直接售卖由此模型生成的图片，除非您对该图片进行了足够程度的人工修改，使其在法律意义上可以被完全判定为您的个人作品。如果您违反本条，所造成的一切法律后果由您个人承担，请恕本人概不负责。

3. 您不能使用该模型故意制作或共享非法或有害的内容传播和输出，请您遵守公序良德，将此模型用于积极正面的用途。

本模型中不包含任何真人素材图像源和训练代码来自互联网，模型仅用于科研兴趣交流