I am providing 23 high-quality illuminated photos of the same person and need a LoRA model for Stable Diffusion XL that I can seamlessly load into Fooocus. The generation should enable me to create social media posts in which the face always appears clearly recognizable, consistent, and realistic—preferably with a neutral expression. Please take care of the complete fine-tuning (LoRA, 16-bit .safetensors) based on SDXL 1.0 or an equivalent weighting. It is important that: • the model can be selected directly in Fooocus, • prompt variation does not noticeably change the appearance or facial features, • skin tones and details appear natural and do not show typical training artifacts (marbling, double eyes, etc.). As delivery, I expect the finished .safetensors file, your training settings (learning rate, steps, seed, etc.), and a short usage snippet that I can use to reproduce the result. As soon as I see five different scenes with the same neutral, realistically rendered face in Fooocus for random prompts, the job is considered complete. Translated with DeepL.com (free version)