This project demonstrates how I used a single selfie as the foundation for a new Flux-LoRA model. By leveraging the Hailuo Image-1 model, I generated 15 diverse training images using the subject-reference feature pointing to that one selfie I had taken with my phone. Varying among different styles, posses and situations, always keeping the 1024x1024 resolution and both the minimal sober backgrounds and lighting effects. I prepared captions for each image in text, compressed all together in a ZIP file, and trained a this fine-tuned model with replicate/fast-flux-trainer:56cb4a64 The result: a personalized model based on creative data augmentation from one selfie.