Version: 1.0 Trigger Word: <code>midriff</code> Model Type: LoRA (Z-Image De-Turbo)MidriffZit is a focused LoRA designed to improve how Z-Image models generate midriff looks where the waistline is visible. The LoRA is trained on a curated Korean girls dataset.<h3 id="training-details">Training Details</h3><ul><li>Dataset theme: Korean fashion/portrait references</li><li>Caption: <code>midriff</code></li><li>Resolution: 512</li><li>Steps: 3000</li><li>Batch size: 1</li><li>Noise scheduler: FlowMatch</li><li>Optimizer: AdamW 8-bit</li><li>UNet trained, text encoder frozen</li><li>Loss: MSE</li><li>Content/style balance: Balanced</li><li>LoRA configuration:<ul><li>Linear rank 32 / Conv rank 16</li><li>Alpha 32 / 16</li></ul></li></ul><h3 id="model-base">Model Base</h3>Optimized for the Z-Image model family, specifically:<ul><li>Base: <code>ostris/Z-Image-De-Turbo</code></li><li>Extras: <code>Tongyi-MAI/Z-Image-Turbo</code></li></ul><h3 id="trigger-word">Trigger Word</h3>Recommended LoRA strength: 1.0<pre><code>midriff</code></pre><h3 id="sampler-prompts">Sampler Prompts</h3><ul><li>"DJ at a night club, fisheye lens, smoke machine, laser lights, holding a drink, midriff"</li><li>"Model showing off a new t-shirt at the beach, a shark jumping in the background, midriff"</li><li>"Guitarist performing on stage, laser lights, punk rock scene, midriff"</li><li>"Studio fashion photo, white background, medium shot, modeling clothing, midriff"</li></ul>

Files with SHA256: 58f35b9facad3b9e21a22f20a16d5e60b1bad2b51b9a6a6e9866213e0c745e77