Pretty Luna

Model Overview & Usage Guide

This LoRA is a specialized identity anchor for the character lunavoss, specifically engineered to maintain extreme facial consistency and high-fidelity skin textures across any SDXL-based generation. The model excels at preserving core identity features—such as her specific hair and facial structure—even across different poses and environments. It is designed to produce natural skin textures with visible pores and subtle blemishes, effectively eliminating the common "AI plastic" or waxy look.

For maximum realism, it is optimized for SDXL 1.0 and works best with checkpoints like CyberRealisticXL or RealVisXL. Always include the trigger word lunavoss in your positive prompt. The recommended LoRA strength is between 0.6 and 0.9; if you notice identity drift or distortion, try a weight of 0.85. For the best technical quality, use a CFG Scale of 5.5 and the DPM++ 2M Karras sampler with 35-40 steps.

Avoid using this model with resolutions lower than 1024x1024, as it was trained specifically for SDXL's required density. It should not be used with a LoRA weight of 1.0 or higher if you see over-saturation or "fried" textures. Additionally, this is an SDXL-exclusive resource and will not function correctly with older Stable Diffusion v1.5 checkpoints.