AI-driven photorealistic style transfer: developing a 3D LUT-based iOS app for photo and video retouching

Abstract

This project addresses a critical technology gap in mobile editing by developing an iOS application integrating AI-driven photorealistic style transfer with comprehensive 3D LUT and media editing capabilities. The integration of AI is fundamentally transforming creative workflows, opening entirely novel artistic avenues previously unimaginable, significantly accelerating editing timelines, and enabling highly personalized artistic expressions that adapt to individual styles and content context. Leveraging an on-device CoreML-powered NLUT model, the application enables users to generate 3D LUTs from style images for rapid and accurate color grading across both photos and videos, overcoming the limitations of existing mobile tools that often lack robust LUT creation and unified media processing.

The application, built using SwiftUI and employing the MVVM architecture, provides a user-friendly interface that caters to both professional creatives and casual users. Professionals benefit from custom LUT creation for consistent branding, while casual users enjoy easy media edits and pre-loaded styles. Key features include the NLUT model for efficient 3D LUT generation, LUT editing tools, and media-specific adjustments, all optimized for mobile performance.

All project objectives were met with the development of a fully functional iOS application demonstrating sophisticated mobile color manipulation. Performance testing validated the NLUT model’s suitability for real-time 3D LUT generation. This application uniquely combines AI-powered style transfer with comprehensive 3D LUT editing for both photos and videos on iOS, expanding creative possibilities for mobile users.