<h1 id="nyanko-by-luka-(w)">Nyanko by LUKA (￣ω￣)</h1>Nyanko is a female character created by LUKA (￣ω￣) @onlinepetclub on <a target="_blank" rel="ugc" href="http://X.com">X.com</a>, known to be friends with Wanwan (Wanko). She is a catgirl with cat ears similar to Nicole Watterson from The Amazing World of Gumball.<hr /><h2 id="tags-to-activate-this-character">Tags to Activate This Character</h2><h3 id="base-character">Base Character</h3><pre><code>ny4nk0, 1girl, solo, catgirl, animal ears, brown eyes, white hair, short hair, low twintails, hair between eyes, hair bobbles, four-leaf_clover_hair_ornament, hair ornament, blush stickers, cat tail,</code></pre><h3 id="default-outfit">Default Outfit</h3><pre><code>pink tank top, pink polka dot on tank top, pink bow on tank top, collarbone, blue skirt,</code></pre><h3 id="maid-outfit">Maid Outfit</h3><pre><code>black maid, puffy short sleeves, white apron, long apron, black bow ornament, white bow neck, white bow behind, white socks, white kneehighs,</code></pre><h3 id="important-notes">⚠️ Important Notes</h3>- Maid outfit: the <code>black bow ornament</code> tag may not appear accurately as in the original image- Tags <code>hair bobbles, four-leaf_clover_hair_ornament</code> may not be stable yet- To get the character's ears to match the original best, set the guidance scale to 2<hr /><h2 id="additional-information">Additional Information</h2>One of the reasons this model was released late is because I was interested in this character, but I had already decided beforehand to make it. However, the number of images for this character was quite limited, so the dataset was compressed to the core.Later, video generation tools became available, so I used them to augment the data before finally completing it.<h3 id="development-plans">Development Plans</h3>Currently (**2025/12/7**), I don't have any plans to continue with this model. However, if you want me to create additional ports for it, I can:Version 1:- ✅ Illustrious (available)- ❌ Pony (not available)- ❌ SD15 (not available)- ❌ Flux (not available)- ❌ Z-Image (not available)<hr /><h2 id="training-process-parameters">Training Process Parameters</h2><h3 id="dataset">Dataset</h3>- Chibi: 4 images- Black t-shirt: 6 images- Maid outfit: 6 images- Daily outfit: 13 images- Other: 1 imageConsistency: ModerateTotal dataset: 30 training images<h3 id="technical-specifications">Technical Specifications</h3>| Parameter | Value ||-----------|-------|| Model | Illustrious 0.1 (default) || Image Repetitions | 1-2 || Steps | ~1900 || Epoch | 19/22 || Loss | 0.0399 and Smooth 0.0614 || Learning Rate (Unet) | 1e-4 || Learning Rate (TE) | 8e-6 || Optimizer | AdamW8bit || Scheduler | cosine_with_restarts || Resolution | 1024x1024 || Batch Size | 2 || Flip Augmentation | No || Min SNR Gamma | 5 || Noise Offset | 0.04 || Clip Skip | 2 || Mixed Precision | bf16 || Cache Latents | Yes || Gradient Checkpointing | No || LR Warmup | 10% (200 steps) || ETA | Approximately 2 hours |

Files with SHA256: f49aad89917eb33e901c9deadcf2a6f97658eb6c98f61a144470bf703e84e669