2026.01.31更新了 Z_Image Base 版本，数据集与 Turbo 版本一致。训练6000步。适合 Z_Image Base 也适合 Z_Image Turbo 大模型使用。实测，似乎比使用 Z_Image Turbo 大模型训练的效果在拟真方面要好一点点。2025.11.30更新了 Z_Image Turbo 版本，数据集与 ILL 版本基本一致。作为首个 ZIT 版本，训练3000步。提示词不必再使用质量咒语。2025.07.27更新了 ILL 版本，同样基于1000张由本系列模型所生成的精致图片训练而成，效果良好。本模型仅供学习及测试而使用，请勿用于其它用途。2024.12.30About this version ( YeStyle )本次更新基于1000张最近一年内由本系列模型所生成的优质AI绘画作品（XL+FLUX），其中有一些训练图片是由本系列未公开的Pony模型所生成的。另外还有几张个人收藏的国风摄影作品，以及几张个人特别钟爱的AI绘画作品，作为灵感的补充元素。——因此将其命名为：YeStyle经过几天的测试，新版本似乎在画质的立体感上有了比较明显的进步。作为一年来尝试AI绘画以及敦煌壁画模型的年终总结，希望这款模型能够为喜欢本系列风格的朋友们以及AI绘画爱好者们提供一些新的乐趣。此外，在测试过程中，我惊喜地又发现了一些能够与本系列模型完美适配的优质宝藏模型，其中包括:1、Portrait ArtDorota_pietrowiak<a target="_blank" rel="ugc" href="https://civarchive.com/models/1050585?modelVersionId=1178842">https://civarchive.com/models/1050585?modelVersionId=1178842</a>2、Movers-Dragons and Maidens-Dragons-Photos_Dragons and Maidens<a target="_blank" rel="ugc" href="https://civarchive.com/models/831817/movers-dragons-and-maidens-dragons-photosdragons-and-maidens">https://civarchive.com/models/831817/movers-dragons-and-maidens-dragons-photosdragons-and-maidens</a>3、水墨女孩<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/0cdc18677cc0425f9dcf3d4616b044e6?from=search&amp;versionUuid=9a19019c24724a659058ec89be496a77">美女世界丨水墨女孩-LoRA-随机游戏攻略-LiblibAI</a>4、水彩插画<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/dae847efaee841478fa5c1e661fda2f0?from=search&amp;versionUuid=7ce39231b2214902b82b8f5dc0c1e24c">WaterColor l 水彩插画-F.1_手绘水彩风格插画-LoRA-叶叶叶叶叶-LiblibAI</a>5、扁平插画<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/1f15d943577c467a96b4e1322f4d4e5b?from=search&amp;versionUuid=43c0476d3fbc40f6ad76530085f3dbc8">F.1_XWA_古风扁平插画中国传统风格-LoRA-曦邬桉-LiblibAI</a>6、中国神话 玄幻风格<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/70cd15a83c3849659c7dfc68ec3644d1?from=search&amp;versionUuid=09ee493496a045cca9a3f5b0378a1457">flux版中国神话玄幻风格丹-LoRA-SD炼丹师忠忠-LiblibAI</a>======================This update is based on 1000 high-quality AI paintings (XL+FLUX) generated by the models in this series within the last year, some of which were trained using images generated by an unpublished Pony model from this series. Additionally, there are several personal collections of Chinese-style photography and a few AI paintings that I particularly favor, serving as supplementary elements of inspiration. — Hence, it is named: YeStyle.After a few days of testing, the new version seems to have made significant improvements in the sense of depth in image quality.As an annual summary of trying AI painting and the Dunhuang mural model over the past year, I hope this model can provide some new enjoyment for friends who appreciate the style of this series and AI painting enthusiasts.Moreover, during the testing process, I was delighted to discover some more high-quality treasure models that perfectly complement the models in this series, including:1、Portrait ArtDorota_pietrowiak<a target="_blank" rel="ugc" href="https://civarchive.com/models/1050585?modelVersionId=1178842">https://civarchive.com/models/1050585?modelVersionId=1178842</a>2、Movers-Dragons and Maidens-Dragons-Photos_Dragons and Maidens<a target="_blank" rel="ugc" href="https://civarchive.com/models/831817/movers-dragons-and-maidens-dragons-photosdragons-and-maidens">https://civarchive.com/models/831817/movers-dragons-and-maidens-dragons-photosdragons-and-maidens</a>3、水墨女孩<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/0cdc18677cc0425f9dcf3d4616b044e6?from=search&amp;versionUuid=9a19019c24724a659058ec89be496a77">美女世界丨水墨女孩-LoRA-随机游戏攻略-LiblibAI</a>4、水彩插画<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/dae847efaee841478fa5c1e661fda2f0?from=search&amp;versionUuid=7ce39231b2214902b82b8f5dc0c1e24c">WaterColor l 水彩插画-F.1_手绘水彩风格插画-LoRA-叶叶叶叶叶-LiblibAI</a>5、扁平插画<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/1f15d943577c467a96b4e1322f4d4e5b?from=search&amp;versionUuid=43c0476d3fbc40f6ad76530085f3dbc8">F.1_XWA_古风扁平插画中国传统风格-LoRA-曦邬桉-LiblibAI</a>6、中国神话 玄幻风格<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/70cd15a83c3849659c7dfc68ec3644d1?from=search&amp;versionUuid=09ee493496a045cca9a3f5b0378a1457">flux版中国神话玄幻风格丹-LoRA-SD炼丹师忠忠-LiblibAI</a>===================此模型能够让人物表现出超逸绝伦，仙袂飘然，环佩玉鸣的道家精神与神仙气质，特别推荐国风爱好者使用。此模型根据一些中国古装剧（敦煌风）的宣传海报训练而来，在表现中国传统工笔画的功能上有绝妙的附魔作用——人脸更干净，更清晰，更美观——更符合中国传统审美。==================2024.8.24. 新增 FLUX版推荐与以下 FLUX lora 组合使用，以获得更为活泼的构图：东游记（<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/a5d10dab05234409905143f2f804a2ff?from=feed">Journey to the Eest l 东游记-FLUX-LoRA-叶叶叶叶叶-LiblibAI</a>）风月无边（<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/b275bf18078b41b38e3dbc40d5b3fead?from=feed">风月无边illustrations-LoRA-JuicyBoy-LiblibAI</a>）J_画风_浮世绘（<a target="_blank" rel="ugc" href="https://www.liblib.art/modelinfo/7a77e15599c84653b7155bd2a0865ffc?from=feed">J_画风_浮世绘&amp;暗黑插画-LoRA-J_-LiblibAI</a>）==================可以通过 FaceDetailer 来修复或增强人脸的表现力，通常来说，第一次出图的效果就非常理想了。在 strength_model = 1 时，此模型可以模拟敦煌壁画的某些艺术特征，比如山、云、雾、飘带、以及人物形象等。在 strength_model = 0.68 左右时，此模型可以在敦煌壁画艺术特征之上表现出更多的人物立体感与真实感。在 strength_model = 0.38 左右时，此模型可以为真实人物提供更干净，更清晰，更美观——更符合中国传统审美的人脸表现力。推荐的 CheckPoint，包括但不限于：<a target="_blank" rel="ugc" href="https://civarchive.com/models/596397/sdxlluanfanghua-or-or">乱芳华</a>，<a target="_blank" rel="ugc" href="https://civarchive.com/models/125634/sdxl-ronghua-or-or">荣华_国风</a>，<a target="_blank" rel="ugc" href="https://civarchive.com/models/471568/ororor">扶摇汉服XL</a>，等等。推荐的 LoRA，包括但不限于：<a target="_blank" rel="ugc" href="https://civarchive.com/models/484356/wudaozi">wudaozi</a>，<a target="_blank" rel="ugc" href="https://civarchive.com/models/438857?modelVersionId=488854">扶摇</a>，等等。推荐的提示词，包括但不限于：请酌情增减。<pre><code>dunhuang murals, dunhuang_murals_style, murals art, 
(1girl/immortal, jewelry, earrings, red lips, long sleeves, chinese clothes, cyan hanfu, hair bun, shawl, ribbon, Phoenix Crown, Biyao, Jade Ornaments, Floral Ornaments, armor, Hair Pin, Hair Comb, Golden Forehead Ornament, Beaded Strands, cleavage, huge chest), fog, clouds, mountain, floating in mid air, floating in sky, statdust, 
(make jianjue/holding the index and middle fingers together with the thumb pressing against the ring finger),</code></pre>当然，此模型也可以与任何其它风格的XL模型混用，实际效果还有待各位的积极返图及反馈。======== ======== ========This model can bring out an extraordinary and unparalleled quality in characters, with an ethereal and flowing celestial aura, and the Taoist spirit of jingling jade pendants, making it particularly recommended for fans of traditional Chinese style.This model is trained based on the promotional posters of some Chinese historical dramas (Dunhuang style) and has a wonderful enchanting effect in representing the function of traditional Chinese fine brush painting – the human face appears cleaner, clearer, and more aesthetically pleasing – more in line with traditional Chinese aesthetic standards.The FaceDetailer can be used to repair or enhance the expressiveness of a human face, and typically, the effect of the first image output is very ideal.At strength_model = 1, this model can simulate certain artistic characteristics of Dunhuang murals, such as mountains, clouds, mist, fluttering ribbons, and human figures.At strength_model ≈ 0.68, this model can add a sense of depth and realism to the characters on top of the artistic features of Dunhuang murals.At strength_model ≈ 0.38, this model can provide a cleaner, clearer, and more aesthetically pleasing – more in line with traditional Chinese aesthetic standards – representation of real human faces.Recommended CheckPoints include but are not limited to: <a target="_blank" rel="ugc" href="https://civarchive.com/models/596397/sdxlluanfanghua-or-or">Luan Fanghua</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/125634/sdxl-ronghua-or-or">Ronghua_Guofeng</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/471568/ororor">Fuyao Hanfu XL</a>, and so on.Recommended LoRA includes but is not limited to: <a target="_blank" rel="ugc" href="https://civarchive.com/models/484356/wudaozi">wudaozi</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/438857?modelVersionId=488854">fuyao_Lora_XL</a>, and so on.Recommended prompt words include but are not limited to: Please adjust accordingly<pre><code>dunhuang murals, dunhuang_murals_style, murals art, 
(1girl/immortal, jewelry, earrings, red lips, long sleeves, chinese clothes, cyan hanfu, hair bun, shawl, ribbon, Phoenix Crown, Biyao, Jade Ornaments, Floral Ornaments, armor, Hair Pin, Hair Comb, Golden Forehead Ornament, Beaded Strands, cleavage, huge chest), fog, clouds, mountain, floating in mid air, floating in sky, statdust, 
(make jianjue/holding the index and middle fingers together with the thumb pressing against the ring finger),</code></pre>Certainly, this model can also be mixed with any other style of XL model; the actual effects are yet to be seen and will depend on everyone’s active sharing of images and feedback.

888_DunHuang_Murals_Style

敦煌壁画风格的 Z_Image Turbo 版本。整体感觉上比 SDXL 出图准确（大部分情况下不必修脸修手），比 FLUX 画面干净（没有莫名其妙的竖纹以及侧边纹路），缺点可能随机性不够，Turbo 版 CFG=1，数值&gt;1的情况下，则画面有劣化的迹象。提示词，不必再堆砌质量咒语，直接描述画面即可。提示词中添加打光照亮面部，则会提高面部的清晰度。比如：<pre><code>光线柔和，从上方照射下来，突出了她的面部和服装的细节。</code></pre>训练数据集与 ILL 版本基本一致，没有添加 FLUX 图片，以防出现 FLUX 图片的通病。

Files with SHA256: b31d882aad2f6893196f1d0dc2c13bd446bf5f21ce2892bf91e527fdb958eebb