该lora的设计理念是使用绘画的设计知识，来打破模型的固有偏好。例如角色偏好在画面中心，设计没有章法，想让画面是平整的却总是画出倾斜的。当前还处于探索阶段，某些效果在一些内容上还不错。不要相信该lora能100%出好图。具体效果请参见我的示例图片。感谢大家下载并使用。如果你喜欢，欢迎关注我。如果有需要，可以私信或者在以下位置联系我。<a target="_blank" rel="ugc" href="https://space.bilibili.com/287850276">aiseaK的个人空间-aiseaK个人主页-哔哩哔哩视频</a><a target="_blank" rel="ugc" href="https://discord.gg/xKc4huA376">https://discord.gg/xKc4huA376</a><hr />V1版本为第一个正式版，在多个构图方面强于测试版本。在<code>cmpctrl</code>后第一tags为主构图，后续的为次构图。混合使用时，最好场景围绕主构图来编辑。V1新增horizontal split composition 水平分割构图。V0.2版本多方面优于V0.1版本，但是在配色和光照上，我认为较弱一些（我的审美上）。<hr />V1 is the first non-test version and is stronger than the test versions in several composition aspects. After cmpctrl, the first tags are the primary composition, and the subsequent ones are secondary compositions. When using them in combination, it is best to edit the scene around the primary composition. V1 adds horizontal split composition.V0.2 is superior to V0.1 in many aspects, but in terms of color scheme and lighting, I think it is somewhat weaker (to my aesthetic taste).<hr />The design philosophy of this LoRA is to use design knowledge from painting to break the model's inherent preferences. For example, characters tend to be placed in the center of the frame, the composition lacks structure, and when you want the image to be level, it always comes out tilted.It is currently still in the exploratory stage, and some effects work fairly well on certain types of content.Don't believe that this LoRA can produce good images 100% of the time.For specific results, please refer to my example images.Thank you all for downloading and using it. If you like it, feel free to follow me.If needed, you can DM me or contact me at the following places.<a target="_blank" rel="ugc" href="https://space.bilibili.com/287850276">aiseaK的个人空间-aiseaK个人主页-哔哩哔哩视频</a><a target="_blank" rel="ugc" href="https://discord.gg/xKc4huA376">https://discord.gg/xKc4huA376</a><hr />触发词： <code>cmpctrl</code>## 一、预计要具备的构图设计类型（可混合使用）1. 前景-中景-远景构图： <code>foreground-midground-background composition</code>作用： 增加构图的纵深，更有距离感。角色往往会被绘制在中景或前景的位置。2. 三分法构图： <code>rule of thirds</code>作用： 让角色不那么靠近正中心。3. 三角形构图： <code>triangular composition</code>作用： 让画面看起来更静止，画面的重心趋于下方。往往出现弯腿或坐下的动作。4. 垂直分割构图： <code>vertical split composition</code>作用： 在画面中添加一些垂直的构成，例如柱子与地面垂直。5. 不对称平衡： <code>asymmetrical balance</code>作用： 让画面元素非对称，例如画面左边与右边不一样。6. 对称构图： <code>symmetrical composition</code>作用： 让画面具备一丝对称的结构。7. 中心构图： <code>center composition</code>作用： 让角色位于画面中心。8. S/曲线构图： <code>s-curve composition</code>作用： 让角色或画面具备一定弧度。9. 对角线构图： <code>diagonal composition</code>作用： 倾斜画面或者让角色趋于对角线上。10. 留白构图： <code>negative space composition</code>作用： 在画面中增加一些非角色的空白区域，这个空白是指没有角色。11. 框景构图： <code>framing composition</code>作用： 添加一些类似窥视的构图结构，一般需要场景中有这些能形成框景的。12. 螺旋构图： <code>golden spiral composition</code>作用： 让画面有动感，往往呈现旋转结构。13. 引导线构图： <code>leading lines composition </code>V0.2版本及以后替换为“visual flow compositio”作用： 让画面具备一个视觉中心，部分物品的线头或者透视，引导视线看向视觉中心。14. 黄金分割构图： <code>golden ratio composition</code>作用： 与三分法构图差不了多少，让确实不那么靠近视觉中心。15. 水平分割构图：horizontal split composition作用：在画面增加一个水平分割线，分开画面上下。<hr />Trigger Word: <code>cmpctrl</code>## I. Intended Composition Design Types (Can Be Mixed)1. Foreground-Midground-Background Composition — <code>foreground-midground-background composition</code>Effect: Adds depth to the composition and a stronger sense of distance. Characters are often placed in the midground or foreground.2. Rule of Thirds — <code>rule of thirds</code>Effect: Keeps the character from being too close to the exact center.3. Triangular Composition — <code>triangular composition</code>Effect: Makes the image feel more static, with the visual weight tending toward the lower part of the frame. Actions such as bent legs or sitting often appear.4. Vertical Split Composition — <code>vertical split composition</code>Effect: Adds some vertical elements to the image, such as columns perpendicular to the ground.5. Asymmetrical Balance — <code>asymmetrical balance</code>Effect: Makes the image asymmetrical.6. Symmetrical Composition — <code>symmetrical composition</code>Effect: Gives the image a slight symmetrical structure.7. Center Composition — <code>center composition</code>Effect: Places the character at the center of the image.8. S-Curve Composition — <code>s-curve composition</code>Effect: Gives the character or the image a certain curvature.9. Diagonal Composition — <code>diagonal composition</code>Effect: Tilts the image or makes the character tend toward a diagonal line.10. Negative Space Composition — <code>negative space composition</code>Effect: Adds some non-character empty areas to the image; “empty” here means without characters.11. Framing Composition — <code>framing composition</code>Effect: Adds a voyeuristic/peeking-like compositional structure.12. Golden Spiral Composition — <code>golden spiral composition</code>Effect: Makes the image feel dynamic, often presenting a rotating structure.13. Leading Lines Composition — <code>leading lines composition</code>In V0.2 and later, replaced with <code>visual flow composition</code>.Effect: Gives the image a visual focal point; the lines or perspective of some objects guide the viewer’s eye toward the focal point.14. Golden Ratio Composition — <code>golden ratio composition</code>Effect: Similar to rule of thirds; keeps the subject from being too close to the visual center.15. Horizontal split composition: horizontal split compositionFunction: Adds a horizontal dividing line to the image, separating the top and bottom of the frame.

构图自控 | Composition Self-Control - 生图设计 | Raw image design

整数的都是正式版，往往将该版本之前的特征融合得到的结果。画风lora与该lora搭配时，该lora效果会有所损失。V1采用与V0.1和0.2不同的训练策略和数据集，单构图提示词能力提升，混合情况下也有较好的结果。单构图虽然提升了，但混合的时候有的构图可能会有冲突。因此我让AI稍微整理了一下如何使用，可以下载附件作为参考使用。Integer-numbered versions are the non-test versions, as opposed to the test versions, and they are usually the result of merging features from previous versions. When a style LoRA is used together with this LoRA, the effect of this LoRA will be somewhat diminished.V1 uses a different training strategy and dataset from V0.1 and V0.2. Its single-composition prompt capability is improved, and it also gives fairly good results in mixed cases.Although single composition has improved, some compositions may conflict when mixed. So I had AI briefly put together how to use it. You can download the attachment and use it as a reference.

Files with SHA256: 5fd0ec12a33fb1800c6a9b7ce6212cd753eda250e429e1188267dec414ecb4af