As of now this 8-Steps <a target="_blank" rel="ugc" href="https://huggingface.co/lightx2v/Qwen-Image-2512-Lightning/blob/main/Qwen-Image-2512-Lightning-8steps-V1.0-fp32.safetensors">LightX2V LoRA produces better output</a> This LoRA trained for a very specific style: close-up portraits featuring platinum blonde / pale silver hair, extremely pale skin with visible freckles and natural texture, and full, plump lips.The training dataset consisted exclusively of tight headshots and facial close-ups , no full-body, half-body, or wider compositions were included.<ul><li>performs best in close-up portrait scenarios ( "close-up face", "portrait", "headshot").</li><li>For anything beyond tight crops (upper body, full body), you should lower the LoRA strength (0.4–0.5 range recommended) to avoid distortions or over-application of the trained features.</li><li>Sampler: euler</li><li>Scheduler: sgm_uniform</li><li>Steps: 10</li><li>CFG: 1</li></ul>

Files with SHA256: 51966489070eaf9e9f54371a9b29386e9ca371aa334a54680d28eaf1d7a3b17b