<p>A character LoRA for <strong>Krea 2</strong>. It locks in one face: a young woman with long dark hair, hazel-green eyes, light freckles and soft natural skin texture. Body, outfit, framing and setting are left entirely to your prompt.</p><h3>How to use</h3><ul><li>Trigger word: <strong>kitty_uc</strong></li><li>Recommended strength: <strong>1.0</strong> - every image in this gallery is at 1.0.</li><li>Gallery settings: CFG 1.0, 13 steps, sampler er_sde / simple scheduler, 2144x928.</li><li>Four of the samples run on stock Krea 2 Turbo, the other six on a Krea 2 community finetune, so it transfers between Krea 2 checkpoints without retuning.</li></ul><h3>Honest note about the samples</h3><p>These images are not kitty_uc alone. They were generated with a stack of public helper LoRAs on top of it - a raw/spicy photo LoRA (0.9), a candid photo LoRA (0.8-1.2), a body-shape LoRA (1.25 on most shots) and a detail/upscale LoRA (0.35-0.6). The face is kitty_uc; the grain, framing and body proportions come from that stack. Used on its own the LoRA gives you the same face with a cleaner, more neutral render.</p><h3>Training</h3><p>Trained on Krea 2 Raw, 2500 steps / 48 epochs, face-focused (masked) training - which is why the identity holds while everything else stays promptable, from a soft snowy portrait to fully explicit scenes.</p><p><strong>Adult content.</strong> Character is synthetic, not a real person.</p><p>Built and tested with <a href="https://github.com/perfectgf/lora-dataset-studio" rel="ugc">LoRA Dataset Studio</a>, an open-source dataset + training front-end.</p>

Files with SHA256: 1cc9697b618d01ce8ae1b26622ed5fc1cda57b7ff17a7737e5da33aae0aaf4a2