<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: YOLOv8n / YOLOv8s (Ultralytics)</li><li>Task: Object detection, single class (<code>eye</code>)</li><li>Training data: 286 self-generated anime and semi-realistic 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/66d17307-846d-4c10-a1d2-b9667fcfaf65/original=true/66d17307-846d-4c10-a1d2-b9667fcfaf65.jpeg" /><h2 id="known-limitations">Known limitations</h2><ul><li>Detection on male character eyes is improved as of v2.0 but still less reliable than on female characters.</li><li>Trained on anime and semi-realistic style images — also tested well on realistic styles, though not extensively validated across every art style.</li><li>Dataset size (286 images as of v2.0) — strong validation metrics, but a larger dataset would improve robustness further.</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>Easiest method: Search "eye" in ComfyUI-Manager's Install Models menu and install directly — no manual download needed.Manual method:<ol><li>Download the model file — Civitai renames files automatically (e.g. <code>animeEyeDetector_v20.pt</code>)</li><li>(Optional) Rename it to something clear, like <code>eye_yolov8n.pt</code> or <code>eye_yolov8s.pt</code> depending on which size you downloaded — this just determines what shows up in the dropdown menu, doesn't affect how it works.</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)

v2.0 - Retrained on an expanded dataset (286 images, up from 212), specifically targeting known weak spots. Updated both Nano and Small models. Fixes: - Close-up/face-only shots (previously failed to detect entirely) - Chibi-style and large cartoony eyes - 2D/flat anime art styles Improvements: - Higher detection confidence across the board, even on previously working cases - Better handling of unusual/extreme eye angles and zoomed-in crops Notes: - Small model performs best on extreme/wild cases and unusual eye styles - Nano model slightly outperforms Small on very tiny or oddly-shaped eyes - Both models still improving on male character detection — not fully solved, but noticeably better than v1

Files with SHA256: 25564ce95f1889ff9397e1c3e9030564e1c3904db6ad43b7c548df3499bec540