deus-ex-machina/novelai-anime-v3-artist-comparison

--- license: apache-2.0 task_categories: - text-to-image language: - en tags: - art - novelai - nai - example - sample - comparison - stable-diffusion - stable-diffusion-xl size_categories: - 10K<n<100K viewer: false --- 2024-01-13 - I have regenerated all the samples with the following changes: - Artist tag in position two. - Replace all underscores in artists tags with spaces when inserting into prompt. - Scale to 5.3 for consistency. - Adjust prompts for better overall quality and consistency. - Generated after NAI fixed a bug where tokens in certain positions were being ignored. This is a repository of 15,000 SFW samples of artist tags generated using the NovelAI v3 anime image model. For those that prefer it, a zip of the images can be found at https://huggingface.co/datasets/deus-ex-machina/novelai-anime-v3-artist-comparison/blob/zip/images.zip These were selected by post count from Danbooru for all artist tags with greater than 94 posts. Sample Prompt and Settings (with merunyaa as the selected artist tag in this case) NOTE: When inserting artist tags into the prompts, I replace all underscores with spaces, as this can effect the results noticeably. Positive ``{patchouli knowledge}, merunyaa, solo, standing, upper body, garden, looking at viewer, facing viewer, pink dress, purple eyes, holding book, best quality, amazing quality, very aesthetic, absurdres`` Negative / Undesired Content ``lowres, {bad}, fewer, extra, missing, worst quality, bad quality, jpeg artifacts, scan artifacts, unfinished, displeasing, chromatic aberration, artistic error, [abstract], @_@, heart-shaped pupils, lineart, sketch, nude, cleavage, panties, nipples, topless`` Settings ``steps: 28, height: 1216, width: 832, scale: 5.3, uncond_scale: 1.0, cfg_rescale: 0.0, seed: 3415329165, n_samples: 1, hide_debug_overlay: False, noise_schedule: native, sampler: k_euler_ancestral, controlnet_strength: 1.0, controlnet_model: None, dynamic_thresholding: False, dynamic_thresholding_percentile: 0.999, dynamic_thresholding_mimic_scale: 10.0, sm: False, sm_dyn: False, skip_cfg_below_sigma: 0.0, lora_unet_weights: None, lora_clip_weights: None``许可证：Apache-2.0 任务类别：文本到图像（text-to-image） 语言：英语（en） 标签：艺术、NovelAI（NovelAI）、nai、示例、样本、对比、Stable Diffusion（Stable Diffusion）、Stable Diffusion XL（Stable Diffusion XL） 样本规模：10000 < 样本数 < 100000 数据集查看器：已禁用 2024年1月13日 —— 本次已重新生成所有样本，更新细节如下： - 将艺术家标签置于提示词第二位 - 将艺术家标签中的下划线替换为空格后插入提示词 - 统一采样比例为5.3以保证生成一致性 - 调整提示词以提升整体生成质量与一致性 - 本次生成于NovelAI（以下简称NAI）修复特定位置Token被忽略的漏洞后完成 本仓库包含15000个全年龄向（Safe For Work，SFW）样本，均使用NovelAI v3动漫图像模型生成，基于艺术家标签构建。 针对有需要的用户，可通过以下链接下载图像压缩包：https://huggingface.co/datasets/deus-ex-machina/novelai-anime-v3-artist-comparison/blob/zip/images.zip 本数据集的样本源自Danbooru平台，选取标准为所有帖子数超过94的艺术家标签。 ### 示例提示词与生成设置（本次示例以merunyaa作为所选艺术家标签） 注意：将艺术家标签插入提示词时，需将其中的下划线替换为空格，该操作会显著影响生成结果。 **正向提示词**： `{帕秋莉知识（patchouli knowledge）}, merunyaa, 单人, 站姿, 半身像, 花园场景, 看向观众, 正对镜头, 粉色连衣裙, 紫瞳, 持书, 最佳质量, 极佳质量, 极具美感, 超高分辨率` **负向/禁用内容**： `低分辨率, {瑕疵}, 数量异常（过少/过多）、缺失, 最差画质, 劣质, JPEG伪影, 扫描伪影, 未完成作品, 观感不佳, 色差, 艺术失误, [抽象风格], @_@, 心形瞳孔, 线稿, 草图, 裸体, 乳沟, 内裤, 乳头, 上身裸露` **生成设置**： `采样步数：28，图像高度：1216，图像宽度：832，采样比例：5.3，无条件引导缩放（uncond_scale）：1.0，CFG重缩放系数：0.0，随机种子：3415329165，单次采样数（n_samples）：1，隐藏调试覆盖层：False，噪声调度方案：native，采样器：k_euler_ancestral，控制网络强度：1.0，控制网络模型：无，动态阈值：False，动态阈值百分位：0.999，动态阈值模拟缩放：10.0，sm：False，sm_dyn：False，低于sigma阈值时跳过CFG：0.0，LoRA UNET权重：无，LoRA CLIP权重：无`

原始信息汇总

数据集更新信息

更新日期

• 2024-01-13

更新内容

• 所有样本已重新生成。