AI-BASED IMAGE TO VIDEO GENERATION SYSTEM

EMENIKE, DANIEL (2026) AI-BASED IMAGE TO VIDEO GENERATION SYSTEM. Other thesis, Godfrey Okoye University, Enugu Nigeria.

[img]
Preview
 Text
( Emenike Daniel ) AI IMAGE TO VIDEO GENERATION SYSTEM FINISHED WORK.pdf
Download (2MB) | Preview

Abstract

In this paper, the use of artificial intelligence in the process of generating video from images was investigated using deep learning algorithms pre-trained by machine learning algorithms. The study aimed at addressing the challenges associated with video generation systems, specifically high computation cost and inconsistent motion in the videos through the development of a hybrid 3D U-Net and diffusion models. The use of a dataset named Panda-70M that comprised video- text clips was done after the frames were extracted, resized, normalized, and filtered. In addition, transfer learning and temporal attention mechanism techniques were used to reduce computational cost and improve consistency of the generated video motion. The performance of the model was analyzed using SSIM (Identity) score of 0.6663, PSNR score of 14.98 dB, Motion Intensity score of 16.87, and Temporal Consistency of 0.0561.The system was able to obtain a mean video generation time of 118.27 seconds per sequence with a frame rate of 0.0676 FPS, indicating that the pretrained diffusion model was able to generate visual video sequences with motion consistency and structural preservation. React Native was utilized for the frontend interface, while Flask and FastAPI were used to implement the backend server handling the communication between the user interface and AI model components. Images along with text prompts can be uploaded by the user to generate motions, and the generated videos can be downloaded automatically by the user. In general, the experiment illustrated the applicability of the pretrained diffusion models for generating videos from images.

Item Type: Thesis (Other)
Subjects: Q Science > Q Science (General)
Divisions: Faculty of Engineering, Science and Mathematics > School of Electronics and Computer Science
Depositing User: Cynthia Ugwuoti
Date Deposited: 22 Jul 2026 14:22
Last Modified: 22 Jul 2026 14:22
URI: http://eprints.gouni.edu.ng/id/eprint/6043

Actions (login required)

View Item View Item
GOUNI Repository is powered by EPrints 3 which is developed by the School of Electronics and Computer Science at the University of Southampton. More information and software credits.