Prepares portrait LoRA training datasets: face-centered 1024x1024 crops, mandatory visual QA and manual crop correction, contact sheets, ZIP packing, and descriptive [trigger] captions via Florence-2. Use when preparing images for FLUX portrait training, fal-ai/flux-lora-portrait-trainer, Kohya, or person LoRA training; or when the user mentions dataset prep, captioning training images, cropping faces for LoRA, or building training ZIP archives.
