<h1 id="horse-pussy-detection-models-8vfzhubjf">Horse Pussy Detection Models</h1>Hello there, fellow mare enjoyers! This is an <a target="_blank" rel="ugc" href="https://docs.ultralytics.com/">Ultralitycs</a> detection model for horse pussy. It can be used for automatic marehood enhancement with the help of special UI extensions (<a target="_blank" rel="ugc" href="https://github.com/Bing-su/adetailer">ADetailer</a> for A1111/[re]Forge, and <a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Subpack">ComfyUI Impact Subpack</a> for ComfyUI, for example). It is done by detecting, masking and inpainting the detected object "at full resolution" (basically, the object is inpainted at low denoising while being "zoomed in", thus adding detail).Note: detection model only detects and masks the object. It does not have any effect on the quality or coherency of the final result, since actual inpainting is usually done with the same checkpoint that produced the full image.There are currently 2 versions of the model:<ul><li>V1 only provides the "bounding box". It masks the entire rectangular area that the horse pussy fits in. It is fairly simple and reliable, but can lead to unwanted side effects, since the entire area is inpainted, not just the horse pussy.</li><li>V2 is a "segmentation" model. It means that it tries to mask only the horse pussy, which improves inpainting quality, but it's not quite as good at detecting edge cases (penetrated pussy, on side, etc.). Hit rate is still pretty good, though. As a bonus, it also tries to detect dock and teats/crotchboobs.</li></ul><h3 id="recommended-v2-adetailer-settings:-99r9wx71h">Recommended V2 ADetailer settings:</h3><ul><li>Detection<ul><li>Detection model confidence threshold: <code>0.35-0.4</code></li></ul></li><li>Mask Preprocessing<ul><li>Mask merge mode: <code>Merge</code></li></ul></li><li>Inpainting<ul><li>Inpaint denoising strength: <code>0.3-0.4</code></li><li>For SDXL-based models:<ul><li>Use separate width/height: <code>True</code></li><li>Width, Height: <code>1024</code></li></ul></li></ul></li></ul>

ADetailer Horse Pussy

Trained on top of yolov11n-seg (segmentation, multiclass).Recommended ADetailer settings:<ul><li>Detection<ul><li>Detection model confidence threshold: <code>0.35-0.4</code></li></ul></li><li>Mask Preprocessing<ul><li>Mask merge mode: <code>Merge</code></li></ul></li><li>Inpainting<ul><li>Inpaint denoising strength: <code>0.3-0.4</code></li><li>For SDXL-based models:<ul><li>Use separate width/height: <code>True</code></li><li>Width, Height: <code>1024</code></li></ul></li></ul></li></ul>

Files with SHA256: 5ac68c9534768a97c55e37ba092c6edbe8ea99ea4b93205449fea1450d0f6f7e