<h1 id="character-lora-maya-v2.1">Character LoRA - Maya v2.1</h1>This LoRA was trained on Illustrious XL and features the character Maya with distinctive pink hair.<h2 id="trigger-words:">Trigger Words:</h2><ul><li><code>maya</code> - Primary character trigger</li><li><code>pink_hair</code> - Hair color/style trigger</li><li><code>long_hair</code> - Hair length control (Added in v1.1)</li><li><code>ponytail</code> - Optional hairstyle variation (Added in v1.1 - recommend "low ponytail" for original style)</li></ul><h2 id="previous-updates">Previous Updates</h2>What's New in v2.0:<ul><li>Closest to original art style yet - Significant improvements over all v1.x versions</li><li>Enhanced preprocessing - Completely revamped training data preprocessing pipeline</li><li>Extended training cycles - Increased training steps with careful manual selection of optimal checkpoints</li><li>Hand-picked model - Each release candidate was manually evaluated for quality and sample consistency</li><li>Maintained compatibility - Tags and base model remain unchanged since v1.1 for seamless upgrading</li></ul>What's New in v1.2:<ul><li>Extended training with more epochs for improved stability</li><li>More consistent generation results</li><li>All v1.1 features remain compatible</li><li>The sample images may appear slightly worse than v1.1 because they are now generated using a much simpler workflow. For the best results, I recommend using waiNSFWIllustrious or other Illustrious models with sophisticated workflows like 2-pass generation.</li></ul>What's New in v1.1:<ul><li>Changed base model to waiNSFWIllustrious_v140.safetensors</li><li>Improved quality of training data</li><li>Standard generation methods should work better now</li><li>Art style now looks more like the original style</li><li>Added new trigger word "long_hair"</li><li>Optionally "ponytail" - if you prefer ponytail, recommend low ponytail to preserve original art style</li></ul><h2 id="important-usage-notes:">Important Usage Notes:</h2><ul><li>Due to the original training images having poor quality, this LoRA may not work well with standard generation methods</li><li>Recommended workflow: Use the embedded ComfyUI workflow included in the sample images for best results</li><li>Load the workflow directly into ComfyUI for optimal generation quality</li></ul>

Maya - The Erotic Temptress [Illustrious]

After some time has passed, I would like to create a new version of this Lora. I think an old one was probably overtrained. This version should clearly be better than all previous versions.

Files with SHA256: 0f13f8951367edb16499da22a350a5eaf68346b0b235a7bc9d18d55ca0b30a7d