AI-Driven Realistic Human Face Creation from Natural Language Prompts

Abstract

: The synthesis of photorealistic human faces from descriptive inputs has become feasible with the advancement of generative artificial intelligence. This paper presents C72: AI-Driven Realistic Human Face Creation, a framework that converts eyewitness free-text descriptions or guided attribute inputs into structured JSON and subsequently generates realistic human faces. The system leverages Google Gemini for attribute parsing and image generation, integrated with a Streamlit interface for interactive use and OpenCV for sketch preprocessing. A deterministic prompt-building process ensures that essential attributes such as age, skin tone, hairstyle, and facial features are preserved across generations, while refinement controls enable variation in pose, lighting, and background. Experimental demonstrations show that the system achieves strong alignment between descriptive attributes and generated outputs, offering potential applications in forensic sketch enhancement, creative industries, and interactive digital media. Ethical safeguards, including consent validation and prevention of public figure resemblance, are incorporated to ensure responsible deployment.