emb.etimen

NST

AI in Digital Art

Exploring generative neural networks for dynamic art composition and interactive digital experiences. This research integrates deep learning with creative workflows to enable real-time style transfer and AI-driven visual installations.

Objective

Our research investigates real-time neural style transfer algorithms implemented with TensorFlow.js and WebGPU for interactive art applications. This project enables artists to apply complex stylistic filters to live video feeds or canvas compositions with sub-second latency through optimized neural processing techniques.

Key Technologies

  • TensorFlow.js for browser-based deep learning
  • WebGPU-powered neural rendering pipeline
  • Real-time video feed integration

Features

  • 60fps style transfer using GPU acceleration
  • Style blending and layer controls
  • Multi-model neural processing pipeline

Style Transfer Demo

Note: This is a conceptual interface. Actual implementation requires WebGL support.

Publication Details

Whitepaper

Technical exploration of style transfer optimization techniques with performance benchmarks comparing different convolutional network architectures.

Read Paper

Codebase

Open-source implementation including trained models, WebAssembly optimizations, and shader code for GPU accelerated computation.

View on GitHub

Demo

Interactive web application demonstrating real-time style transfer on live video input without external dependencies.

Open Demo