<h2 id="breast!">Breast!</h2>Breast! LoRA is here to restore proportions as the cornerstone for the beautiful and the sacred.<h3 id="specification:">Specification:</h3><ul><li>Base Model: SDXL 1.0</li><li>Type: LoRA</li><li>Training Steps: 900 steps, 71 images total, masked and captioned individually by me.</li><li>File Size: 200 MB</li></ul><h3 id="description">Description</h3>We all went through the"bigger is better" era, and we were far from being wrong. The fact is that i felt that we had not yet created a series of LoRAs that would somehow regard proportions between the breast and the feminine body.<h3 id="technical">Technical</h3><span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px">Images are preprocessed using a pose estimation model or a semantic segmentation network to extract keypoints/masks for <code>torso</code><span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px">, <code>waist</code><span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px">, and <code>breast</code><span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px"> regions. For each image ( I_i ), computed isolate perceptions for Torso-to-waist, Breast-to-Torso and Waist-to-Breast ratios, that were then represented in a low dimensional folder structure.<span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px">The whole idea was to make sure that the U-Net’s cross-attention layers are conditioned on text prompts in such a way that we effectively disentangle anatomical traits from non-useful textual semantics<span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px">.<span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px">This is my first LoRA, and the methodology tries and explore the sparsity in anatomical variation to create interactive and compartmentalized LoRA adapters, effectively acting as a form of mixture-of-experts fine-tuning<span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px"> for diffusion models.<span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px">For breast size specifically, I've used "large breasts:yy" tag to teach the diffusion net about sizes, and they pair with a denotative tag about the size, that ranges from large to gigantic:<pre><code>large (yy=0.5 to 0.7), huge (0.8 to 1.2), massive (1.2 to 1.5), gigantic (1.5 and forth).</code></pre><pre><code>Example: ... ((massive breasts, large breast:1.3)) ...</code></pre><span style="background-color:rgb(27, 27, 30);color:rgba(255, 255, 255, 0.8);font-family:plus-jakarta-sans, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, Oxygen, Ubuntu, Cantarell, &quot;Fira Sans&quot;, &quot;Droid Sans&quot;, &quot;Helvetica Neue&quot;, sans-serif;font-size:14px">Each one of them having their own softness and size parametric behavior.What Makes It Special:<ul><li>Decoupled Aging: Breasts remain anatomically correct without affecting facial age.</li><li>Face Preservation: Minimal facial distortion even at higher weights (tested up to 1.9).</li><li>Shape Control: Precise slider-like behavior through trigger words.</li><li>Clothing Interaction: Realistic fabric deformation and layering.</li><li>Multi-Style Compatible: Works with illustration, anime, and realistic photography.</li></ul><h3 id="trigger-words-system">Trigger Words System</h3><pre><code>**Size Control:**
[small, medium,] large (0.5 to 0.7), huge (0.8 to 1.2), massive (1.2 to 1.5), gigantic (1.5 and forth).

**Shape &amp; Form:**
round, full, natural, sagging, tear

**Detail Enhancement:**
backboob, underboob, sideboob,

**Pose/Clothing:**
grabbing, squeezing, hiding
</code></pre><h3 id="training-dataset">Training Dataset</h3><ul><li>Total Images: 71 curated, tagged and masked images.</li><li>Resolution: 1024x1024</li><li>Base model: SDXL 1.0</li><li>AI Training Tool</li><li>Home trained with a RTX 5070.</li></ul><h3 id="usage-instructions">Usage Instructions</h3>Recommended Settings:<ul><li>Weight Range: 0.5-0.8 (optimal), up to 1.2 for extreme results</li><li>CLIP Skip: 2 (for SDXL base models)</li><li>Sampler: Euler A or DPM++ 2M SDE Karras</li><li>HiRes Fix: Compatible at 0.4-0.6 denoising strength</li></ul>Common Pitfalls:<ul><li>Weights above 1.5 cause anatomical distortions</li><li>Avoid combining with [specific incompatible LoRAs]</li><li>For face preservation, use weight ≤0.8</li></ul><h3 id="support:">Support:</h3><ul><li>⭐⭐⭐⭐⭐ ratings help development</li><li>Please do not distribute outside CivitAI</li><li>Hit me in DMs for comments or to have a chat :)</li><li>Report issues via CivitAI comments</li><li>Follow for update notifications</li></ul><h3 id="version-history">Version History</h3>v.1 - 27.12.2025

Files with SHA256: e6a38ffec8b11b8fa4f57e64daeed8185787156536444e3c234a9625090906b1