<h3 id="model-introduction"><strong>Model Introduction</strong></h3><p>This is a test model that has not undergone in-depth quality validation. Feedback is welcome if any issues arise.</p><p>This model was trained using the Concept Slider mode in AI Toolkit, without relying on dataset content, aiming to verify whether strongly stylized LoRAs can be trained using the model's prior knowledge. The model is trained on the concept axis of Realistic Photography ↔ Geometric Composition.</p><h3 id="function-description"><strong>Function Description</strong></h3><p>Adjust the LoRA weight to achieve a continuous transition between realistic photography and geometric constructivism.</p><p><strong>Positive weights</strong>: Tend toward Geometric Constructivism. The image is organized by a geometric skeleton, forms are simplified into geometric blocks, color areas are divided by hard edges without gradients, flat rendering, composition based on horizontal, vertical, and diagonal grids, abstract geometric forms run through the image, emphasizing a sense of order and structure.</p><p><strong>Negative weights</strong>: Tend toward Realistic Photography. Three-dimensional lighting and shadows, soft gradient transitions, depth of field blur, natural material textures, organic forms, strong volumetric feel, naturalistic style.</p><h3 id="usage-recommendations"><strong>Usage Recommendations</strong></h3><p>Example images were tested with weights from -10.0 to 30.0. The higher the positive weight, the stronger the geometric composition style; negative weights maintain the realistic photography direction.</p><p>It is recommended to start with 10.0; weights of 20.0 to 30.0 can achieve a noticeably strong stylization effect.</p><p>No specific trigger words are required; simply adjust the LoRA weight to take effect.</p><h3 id="test-nature-statement"><strong>Test Nature Statement</strong></h3><p>This model is a proof-of-concept test and has not undergone in-depth quality validation. Issues such as style instability, image corruption, or non-linear weight response may exist. If you have any questions or suggestions, feedback is welcome.</p><p></p><h3><strong>模型简介</strong></h3><p>这是一个<strong>测试模型</strong>，并未经过深度的质量验证，如果有问题欢迎反馈。<br />本模型使用 <strong>AI Toolkit 的 Concept Slider 模式</strong>训练，<strong>不依赖数据集内容</strong>，旨在验证能否依靠模型先验知识来训练强风格化 LoRA。模型以<strong>写实摄影 ↔ 几何构成</strong>为概念轴训练。</p><h3><strong>功能说明</strong></h3><p>调节 LoRA 权重即可在写实摄影与几何构成主义之间连续过渡。</p><p><strong>正向权重</strong>：趋向几何构成主义。画面由几何骨架组织，形态简化为几何体块，色彩区域硬边分割无渐变，扁平化渲染，构图基于水平垂直对角线网格，抽象几何形态贯穿画面，强调秩序感与结构感。</p><p><strong>负向权重</strong>：趋向写实摄影。三维立体光影，柔和渐变过渡，景深虚化，自然材质纹理，有机形态，体积感强，自然主义风格。</p><h3><strong>使用建议</strong></h3><p>例图测试权重为 <code>-10.0</code> ～ <code>30.0</code>。正向权重越高，几何构成风格越强烈；负向权重保持写实摄影方向。</p><p>建议从 <code>10.0</code> 开始尝试，<code>20.0</code> ～ <code>30.0</code> 可获得明显强风格化效果。</p><p>无需特定触发词，直接调节 LoRA 权重即可生效。</p><h3><strong>测试性质说明</strong></h3><p>本模型为概念验证性测试，未经过深度质量验证，可能存在风格不稳定、画面崩坏或权重响应非线性等问题。如有任何问题或建议，欢迎反馈。</p>

Files with SHA256: dcf397a2bf6e8d194f8860245f117e20df7e7e5dbe3c4362561b98efa92ad031