Image-to-Video Generative Modeling for Controllable Image Animation

This project aims to develop an AI-driven system that transforms a single still image into a realistic, controllable animated video. Unlike traditional video production, which requires costly filming or manual animation, our approach leverages advanced diffusion models to automatically generate dynamic motion while preserving the original visual details of both characters and text. A key innovation is the creation of layered video outputs, where different elements—such as background, characters, and text overlays—are separated. This allows easy editing and customization, enabling creators to modify individual components without regenerating the entire video.

The technology directly addresses challenges faced by the digital marketing and creative media industries, where rapid production and personalization are essential. By providing a scalable solution for high-quality animation, the project reduces production costs and turnaround times, empowering companies to deliver engaging content across multiple campaigns and markets.

Beyond commercial applications, this work contributes to advancing Canada’s AI and creative technology ecosystem. It trains highly skilled talent in generative AI and fosters collaboration between academia and industry. The outcome will support small businesses, independent artists, and marketing teams with accessible, professional-grade animation tools, driving innovation in digital storytelling and interactive advertising.

Faculty Supervisor:

Konstantinos Plataniotis

Student:

Partner:

Kyoso

Discipline:

Computer science

Sector:

Information and cultural industries

University:

University of Toronto

Program: