<h1 id="anime-eye-detector-(yolov8)">Anime Eye Detector (YOLOv8)</h1><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3d89e1a9-f5a8-4185-8ff0-d884aad3b771/original=true/3d89e1a9-f5a8-4185-8ff0-d884aad3b771.jpeg" />Also available on Hugging Face🤗: <a target="_blank" rel="ugc" href="https://huggingface.co/killjoyelite/anime-eye-yolov8">https://huggingface.co/killjoyelite/anime-eye-yolov8</a>A YOLOv8 object detection model fine-tuned to detect eyes in anime-style character art, intended for use with <a target="_blank" rel="ugc" href="https://github.com/comfyanonymous/ComfyUI">ComfyUI</a> + <a target="_blank" rel="ugc" href="https://github.com/ltdrdata/ComfyUI-Impact-Pack">Impact Pack</a> for automated eye detailing/inpainting workflows (similar to how <code>face_yolov8n.pt</code> and <code>hand_yolov8n.pt</code> are used).<h2 id="model-details">Model details</h2><ul><li>Base model: <code>yolov8n.pt/yolov8s.pt</code> (Ultralytics)</li><li>Task: Object detection, single class (<code>eye</code>)</li><li>Training data: 212 self-generated anime-style images (AI-generated, primarily female characters), manually labeled with bounding boxes around each visible eye</li><li>Training config: 100 epochs, image size 640, batch size 8</li></ul><h2 id="performance-(on-validation-split)">Performance (on validation split)</h2><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/417d8775-aa62-480f-a05d-0ac00b5b98b9/original=true/417d8775-aa62-480f-a05d-0ac00b5b98b9.jpeg" /><h2 id="known-limitations">Known limitations</h2><ul><li>Trained predominantly on female anime characters — detection on male character eyes is less reliable and may miss detections.</li><li>Struggles with very large, cartoony/chibi-style eyes that deviate significantly from standard anime proportions or sometimes closeup of faces (WIP).</li><li>Trained entirely on a single generation style/checkpoint's output — may generalize less well to very different art styles (e.g. heavily stylized, painterly, or non-anime art) than to mainstream anime/semi-realistic anime styles.</li><li>Small dataset (212 images) — while validation metrics are strong, real-world robustness across the full diversity of anime art is inherently more limited than a larger, more varied dataset would provide.</li></ul>If you find specific failure cases, feel free to open a discussion — this is a good candidate for community-driven dataset expansion over time.<h2 id="examples">Examples</h2>Detection preview — the model correctly finds eyes across different poses/styles:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/000c1bdf-610e-43bd-8f9b-0663db2cd018/original=true/000c1bdf-610e-43bd-8f9b-0663db2cd018.jpeg" />Eye color change/Eye fixing<span style="background-color:rgb(16, 22, 35);color:rgb(179, 188, 201);font-family:&quot;Source Sans Pro&quot;, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, &quot;Helvetica Neue&quot;, &quot;Noto Sans&quot;, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;, &quot;Segoe UI Symbol&quot;, &quot;Noto Color Emoji&quot;;font-size:16.8px"> — using the detected eye region with <code>Detailer (SEGS)</code><span style="background-color:rgb(16, 22, 35);color:rgb(179, 188, 201);font-family:&quot;Source Sans Pro&quot;, -apple-system, BlinkMacSystemFont, &quot;Segoe UI&quot;, Roboto, &quot;Helvetica Neue&quot;, &quot;Noto Sans&quot;, Arial, sans-serif, &quot;Apple Color Emoji&quot;, &quot;Segoe UI Emoji&quot;, &quot;Segoe UI Symbol&quot;, &quot;Noto Color Emoji&quot;;font-size:16.8px"> to redraw eye color/detail from a prompt, while keeping the rest of the image untouched:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/ba2520d7-a3cd-4b62-a7cc-b0360bf6e6da/original=true/ba2520d7-a3cd-4b62-a7cc-b0360bf6e6da.jpeg" /><h2 id="usage-(comfyui)">Usage (ComfyUI)</h2><ol><li>Download the model file — Civitai renames files automatically (e.g. <code>animeEyeDetector_v10.pt</code>, <code>animeEyeDetector_v11Small.pt</code>).</li><li>Place it in:<pre><code>ComfyUI/models/ultralytics/bbox/
</code></pre></li><li>Restart ComfyUI.</li><li>In your workflow:<pre><code>Load Image → UltralyticsDetectorProvider (select the eye detector model) → BboxDetectorSEGS → Detailer (SEGS)
</code></pre></li><li>Recommended <code>Detailer (SEGS)</code> starting settings for eye detailing:<ul><li><code>guide_size</code>: 512</li><li><code>denoise</code>: 0.5–0.7 (lower = closer to the original eye, higher = more prompt-driven reinterpretation)</li><li><code>feather</code>: 5–10</li></ul></li></ol><h2 id="license">License</h2>Released under the MIT License. Training images were self-generated by the author; users should independently verify licensing terms of any base checkpoint used to generate their own training/inference images if that matters for their use case.

Anime Eye Detector (YOLOv8)

v1.1 - Added Small version. YOLOv8s base. mAP50: 0.994, mAP50-95: 0.705. Picks up distant, bad quality eyes better than Nano version in testing. Consistently better detection at higher threshold.

Files with SHA256: 20a27ec467a23d6c4b1d7e4859a18cd075ee21d097e7d4400a0e7267f3d285fb