Implementasi Model L-CogVideoX untuk Pembangkitan Video Iklan Minuman Kemasan

Robbani, Maulana Yusuf Ikhsan (2025) Implementasi Model L-CogVideoX untuk Pembangkitan Video Iklan Minuman Kemasan. Diploma thesis, Institut Teknologi Sepuluh Nopember.

Text 5002211167-Undergraduate_Thesis.pdf - Accepted Version Restricted to Repository staff only Download (12MB) | Request a copy

Abstract

Video iklan minuman menjadi kategori periklanan dengan nilai pasar terbesar di dunia. Sayangnya, dalam pembuatan video periklanan secara konvensional ditemukan bahwa biaya produksi video periklanan masih mahal. Keberadaan model Generative Artificial Intelligence bernama L-CogVideoX dapat mengurangi biaya produksi video periklanan. Penelitian Tugas Akhir ini bertujuan untuk melatih sebuah model pembangkitan video, L-CogVideoX dengan masukan deskripsi video agar model dapat menghasilkan video iklan minuman. Adapun model text-to-video CogVideoX memiliki keterbatasan menghasilkan video dengan domain video tidak berfokus pada iklan minuman dan parameter yang sangat besar. Oleh sebab itu, penelitian Tugas Akhir ini mengusulkan sebuah model text-to-video dengan mengintegrasikan konsep Low-Rank Adaptation yang bernama L-CogVideoX. Penelitian Tugas Akhir ini menawarkan
dataset video iklan minuman kemasan untuk model text-to-video. Hasil eksperimen menunjukkan bahwa model L-CogVideoX unggul pada enam metrik dengan skor total sebesar 0,6720 dibandingkan model baseline CogVideoX mencapai 0,6696.

=====================================================================================================================================

Beverage advertising videos are the advertising category with the largest market value in the world. Unfortunately, in conventional advertising video production, it has
been found that the cost of producing advertising videos is still high. The existence of a Generative Artificial Intelligence model called L-CogVideoX can reduce the cost of
producing advertising videos. Penelitian Tugas Akhir aims to train a video generation model, L-CogVideoX, with video description input so that the model can produce beverage advertising videos. The text-to-video model CogVideoX has limitations in generating videos with domains not focused on beverage advertisements and very large parameters. Therefore, this study proposes a text-to-video model by integrating the Low-Rank Adaptation concept named L-CogVideoX. This paper offers a dataset of beverage advertisement videos for the text-to-video model. Experimental results show that the L-CogVideoX model outperforms the baseline CogVideoX model on six
metrics, achieving a total score of 0.6720 compared to 0.6696.

Item Type:	Thesis (Diploma)
Uncontrolled Keywords:	Text-to-video, Video Iklan Minuman, Model Difusi, Generative AI, Text-to-Video, Beverage Advertising Video, Diffusion Model, Generative AI
Subjects:	Q Science > QA Mathematics > QA336 Artificial Intelligence
Divisions:	Faculty of Science and Data Analytics (SCIENTICS) > Mathematics > 44201-(S1) Undergraduate Thesis
Depositing User:	Maulana Yusuf Ikhsan Robbani
Date Deposited:	01 Aug 2025 07:48
Last Modified:	01 Aug 2025 07:48
URI:	http://repository.its.ac.id/id/eprint/125782

Actions (login required)

View Item