Magic Portrait: generative AI that keeps the face

AI art tools quietly “beautify” people — they smooth, idealize and change the face. For a portrait you would hang on a wall, that is the one thing you cannot do. Magic Portrait had to turn any photo — pets, family, kids, couples — into handmade-looking art while preserving identity perfectly, and compose several people into one cohesive piece.

I designed the product around a structured, versioned prompt system — a single source of truth built from composable blocks: an identity-lock rule that forbids reshaping faces, per-technique style blocks, and mood presets (Royal, Heritage, Neoclassical). Gemini 2.5 Flash generates from several photos at once, detecting whether they are the same person or a group and composing accordingly. Generation is async with status polling, previews are watermarked, and checkout reuses the same Medusa + Stripe backend as Portraits.