<h2 id="krea-ashley">Krea Ashley</h2>A photoreal character LoRA for Krea 2: Ashley, a Korean woman with soft ulzzang makeup. She keeps the same face, hair and look across portraits, outfits, places and lighting, from café snapshots to night neon and bedroom shots.<h3 id="how-to-use">✨ How to use</h3><ul><li>Trigger: <code>Ashley</code></li><li>Start every prompt with: <code>Ashley, a young woman in her twenties,</code> then describe the outfit, pose, place and light.</li><li>Don't describe her face or hair: the LoRA brings them. Hairstyles (bun, ponytail, wet hair) are fine.</li><li>Strength: 1.0 to 1.25 (the samples use 1.25). Lower it a little if it fights a style LoRA.</li><li>Models: trained on Krea 2 Raw, made for the Turbo models: <a target="_blank" rel="ugc" href="/models/2760803?modelVersionId=3151280">DaSiWa Krea2 Turbo</a> or <a target="_blank" rel="ugc" href="/models/573152?modelVersionId=3112728">Lustify</a>.</li><li>Settings: 8 steps, CFG 1, er_sde / simple, about 1 MP (896x1152, 832x1248, 1248x832).</li><li>Framing: portrait to thigh-up works best; in far full-body shots her face gets small and loses detail.</li></ul><h3 id="example-prompt">📝 Example prompt</h3><blockquote>Ashley, a young woman in her twenties, wearing an oversized beige trench coat over a white t-shirt, walking down a quiet city street at dusk and glancing back over her shoulder at the viewer, upper body shot, warm shop lights on her face and a cool blue evening sky behind her, natural skin texture, 35mm film look</blockquote><h3 id="the-quality-stack-in-the-samples">🎨 The quality stack in the samples</h3>Each comparison shows left = Ashley alone and right = Ashley + this stack (same seed):<ul><li><a target="_blank" rel="ugc" href="/models/2775340?modelVersionId=3125118">Krea2 TextFusion Refusal-Reduction</a> 0.75</li><li><a target="_blank" rel="ugc" href="/models/929497?modelVersionId=3077110">Aesthetic Quality Modifiers - Masterpiece</a> 1.0</li><li><a target="_blank" rel="ugc" href="/models/2730896?modelVersionId=3070143">Krea2_Enhancer</a> 0.6</li><li><a target="_blank" rel="ugc" href="/models/2729908?modelVersionId=3068874">[KREA 2] Detail Slider</a> 0.55</li><li><a target="_blank" rel="ugc" href="/models/1883576?modelVersionId=3132463">Octane Render</a> 0.85</li></ul>Thanks to the creators of these LoRAs.<h3 id="training">⚙️ Training</h3>59 images at 1024 px, 944 steps (16 epochs), rank 32 / alpha 32, AdamW 8-bit, learning rate 1e-4, trained on Krea 2 Raw with musubi-tuner.<h3 id="license">📜 License</h3>Krea 2 is licensed under the Krea 2 Community License Agreement. This LoRA is a modified version (fine-tune) of Krea 2 and is not an official Krea product.Built with Qwen: the training images were refined with Qwen-Image 2.1 (Qwen Research License, non-commercial), so this LoRA is shared for free, non-commercial use.Ashley is an original AI character and an adult. Do not use this LoRA to make images of minors or of real people.

Files with SHA256: 42f929484ba4363e92129bffa1b8d4c09e78899773663d756ed32fbe26aa9118