Heterochromia v2 — Odd Eye LoRA for Illustrious/NoobAIAdds heterochromia (odd eyes) to characters with controllable eye colors. Version 2 features expanded color variety and higher resolution training over v1.Trigger: <code>heterochromia</code>Recommended weight: 0.7 – 1.0Color control:<pre><code>heterochromia, left eye red, right eye blue
</code></pre>For stubborn colors boost weight on the color tags:<pre><code>(heterochromia:1.2), (left eye red:1.4), (right eye blue:1.4)
</code></pre>Notes:<ul><li>Without color specification the LoRA will randomly select colors, favoring reds and yellows</li><li>Left/right assignment can occasionally be swapped — reverse the tags if needed</li><li>Works best on close-up and portrait shots</li><li>Pairs well with eye_focus LoRAs at reduced weight (0.5-0.7) to avoid color competition</li><li>Tested on Illustrious and NoobAI-XL architecture</li></ul>Base model: Illustrious XL

Hetrochromia_v2

Dataset:<ul><li>v1: 284 images (72 source × ~4 img2img variants each)</li><li>v2: 105 images, cleaner and more focused</li></ul>Caption format:<ul><li>v1: <code>heterochromia, orange eye, ice blue eye</code> — no directional control, base trigger only</li><li>v2: <code>heterochromia, heterochromia_iridum, left_eye_orange, right_eye_blue, different_colored_eyes</code> — full directional control, multiple triggers</li></ul>Key improvements:<ul><li>Added <code>heterochromia_iridum</code> as secondary trigger</li><li>Added <code>different_colored_eyes</code> as tertiary trigger</li><li><code>left_eye_X</code> and <code>right_eye_X</code> teach the model which color goes on which side</li><li>Red sclera format cleaned up and standardized</li><li>Underscores in all trigger words for cleaner token separation</li></ul>Training settings:<ul><li>v1: 15 repeats, 15 epochs = 31,950 steps, ~17 hours</li><li>v2: 10 repeats, 10 epochs = 5,250 steps, ~3.5 hours</li><li>v2 loss finished at 0.109 vs v1's 0.117 — slightly better convergence</li></ul>What v2 should do better:<ul><li>Consistent left/right eye color assignment when prompted</li><li>Stronger response to specific color combinations</li><li>More reliable red sclera activation</li><li>Better generalization due to cleaner captions</li></ul>v2 LoRA trigger words:Main triggers (always include):<pre><code>heterochromia, heterochromia_iridum, different_colored_eyes
</code></pre>Directional eye control:<pre><code>left_eye_[color], right_eye_[color]
</code></pre>Color options used in training:<pre><code>violet, cyan, silver, green, amber, emerald, sapphire, 
turquoise, teal, pink, burgundy, rose, yellow, purple, 
white, ruby, red, magenta, lime, jade, grey, gold, crimson, black
</code></pre>Red sclera format:<pre><code>left_eye_red_sclera_[iris color]_iris
right_eye_red_sclera_[iris color]_iris
</code></pre>Full example prompts:<pre><code>heterochromia, heterochromia_iridum, left_eye_blue, right_eye_red, different_colored_eyes, 1girl, white hair

heterochromia, heterochromia_iridum, left_eye_red_sclera_silver_iris, right_eye_violet, different_colored_eyes, 1girl, silver hair, gothic outfit

heterochromia, heterochromia_iridum, left_eye_amber, right_eye_cyan, different_colored_eyes, 1girl
</code></pre>Start with LoRA weight 0.7-0.8 and adjust from there.

Files with SHA256: 78121a180d731ccf9e77ff0e776e1cdc841d56e661797f59bfc6906c70a9a1ca