Patent No. 12,334,116 | UCF Ref. No. 2024-005-02
Issued June 17, 2025
The University of Central Florida proudly announces that a new U.S. patent has been issued to UCF researchers Dr. Nazmul Karim, Dr. Nazanin Rahnavard, Dr. Umar Khalid, and Dr. Chen Chen for their invention titled, “Image Diffusion Framework for Text-Guided Video Editing.”
This cutting-edge innovation presents a novel method for adapting text-to-image (T2I) diffusion models for efficient and precise video editing. By leveraging spectral decomposition, the invention enables controlled spectral shifts in the model’s weights, allowing for specific modifications to a video based on a text prompt—without altering the overall visual context.
The framework maintains constant singular vectors while selectively modifying singular values in response to user prompts. A spectral shift regularizer further enhances the model’s stability by limiting changes to the largest singular values, preserving the integrity of the original model while enabling fast and resource-efficient adaptation.
Ideal for real-time or computationally constrained environments such as streaming platforms or interactive media applications, this approach represents a significant advancement in AI-driven content creation.
Please join us in congratulating the inventors on this exciting achievement!