How AI headshot tools use LoRA and your selfies to generate portraits

AI headshot services follow a common pipeline: users upload 10–20 selfies, which are cropped, aligned, and captioned before a lightweight model adapter is trained on them. The key technology is LoRA (Low-Rank Adaptation), which fine-tunes only a small fraction of a base model's parameters, making the process fast and storage-efficient. The adapter's 'rank' setting determines how much detail it can learn — too low and results look generic, too high and the model memorises the training images rather than learning the face. A major pitfall called overfitting occurs when uploaded photos are too similar, causing the model to associate the person's identity with incidental details like lighting or clothing. Providing diverse selfies taken across different days, settings, and angles consistently produces better and more versatile results than submitting many near-identical images.