SYNTHETIC HUMAN TWIN: AN AI - POWERED BEHAVIORAL REPLICATION SYSTEM

Abstract: Facial expression manipulation using artificial intelligence has gained significant attention in recent years due to its applications in digital media, content creation, and virtual identity systems. However, existing AI-based image editing techniques often suffer from the problem of identity drift, where the generated output alters key facial features such as skin tone, eye shape, or facial structure, resulting in images that no longer resemble the original subject. This paper introduces Image Replica (Expression AI), a prompt-driven AI system designed to transform facial expressions in portrait images while preserving the subject’s identity.

The proposed system integrates a modern full-stack web architecture consisting of a Next.js frontend and a Fast API backend, combined with advanced generative AI models. A novel Identity-Preserving Prompt Engine is implemented to enrich user prompts with structural and photorealistic constraints before being processed by diffusion-based image synthesis models. The system utilizes Google's Gemini model for prompt interpretation and OpenAI’s DALL-E model for high-quality image generation.

To ensure privacy, the platform adopts an in-memory processing pipeline that prevents any persistent storage of user images. Experimental results demonstrate that the proposed approach produces highly photorealistic expression transformations while maintaining strong identity consistency across generated outputs. The system provides a scalable, accessible, and privacy-conscious solution for AI-driven facial expression synthesis.

Keywords: Generative Artificial Intelligence, Facial Expression Synthesis, Identity Preservation, Prompt Engineering, Diffusion Models, Image-to-Image Transformation, Privacy-Preserving AI, Full-Stack AI Architecture, Digital Twin Generation.