Streamlining Virtual KOL Generation Through Modular Generative AI Architecture

Tan-Hiep To 1,2
 Duy-Khang Nguyen 1,2
 Minh-Triet Tran 1,2
 Trung-Nghia Le 1,2
1 Faculty of Information Technology, University of Science, Ho Chi Minh City 2 Vietnam National University, Ho Chi Minh City, Vietnam
Paper Code Comming Soon Demo

Abstract

Key Opinion Leaders (KOLs) play a crucial role in modern marketing by shaping consumer perceptions and enhancing brand credibility. However, collaborating with human KOLs often involves high costs and logistical challenges. To address this, we present GenKOL, an interactive system that empowers marketing professionals to efficiently generate high-quality virtual KOL images using generative AI. GenKOL enables users to dynamically compose promotional visuals through an intuitive interface that integrates multiple AI capabilities, including garment generation, makeup transfer, background synthesis, and hair editing. These capabilities are implemented as modular, interchangeable services that can be deployed flexibly on local machines or in the cloud. This modular architecture ensures adaptability across diverse use cases and computational environments. Our system can significantly streamline the production of branded content, lowering costs and accelerating marketing workflows through scalable virtual KOL creation.

GenKOL Features

Workflow of Proposed GenKOL System

Image0

Method

We develop a deep learning system for developing virtual KOL, called GenKOL to assist users in creating images of Key Opinion Leaders (KOLs) for product advertising. This system builds upon our Virtual Try-on technology, allowing users to change outfits, apply various makeup styles, modify image backgrounds to showcase diverse locations, and interact with a range of objects, such as apples and guitars. We propose an architectural model designed to facilitate the integration of AI models as plugins. This architecture allows users to quickly and efficiently install, modify, and add services while conserving resources. Additionally, users can choose freely from the available algorithms, each with its advantages and disadvantages. We also develop the GenKOL System based on this architecture.
Image0

Modular Extensibility

Overview of the plugin-based integration workflow in GenKOL. The system allows new AI models to be registered, mapped to service logic, and executed via a centralized Engine. Dependencies between plugins are managed through a matrix to ensure correct execution order within a pipeline.

Image0

GenKOL Virtual KOL Image Results

Our results with GenKOL. Given an original image (top row, left) and corresponding prompts for each attribute (garment, makeup style, and interaction object), our GenKOL system (rightmost column) synthesizes realistic virtual KOLs that seamlessly combine all elements, including the specified background. Our method supports complex, compositional image generation tasks—such as applying specific garments and makeup styles, adding interactive objects, and adapting the background according to the given prompt—while faithfully preserving the subject's structure and appearance. This demonstrates GenKOL's capability for zero-shot, prompt-guided customization, enabling diverse, flexible, and practical KOL image creation for various marketing scenarios.

Image0

Comparison with ChatGPT 4.0 and Gemini 2.0

We evaluated the GenKOL system alongside some of the latest models for image generation in a controlled comparison against top rated image AI models , namely Gemini-2.0-Flash-Preview-Image-Generation and ChatGPT 4o in terms of image generation time, image quality, and consistency.

Image0
GenKOL Features

Citation Information

@inproceedings{
  hiep2025streamlining,
  title={Streamlining Virtual {KOL} Generation Through Modular Generative {AI} Architecture},
  author={To Tan Hiep and Duy-Khang Nguyen and Minh-Triet Tran and Trung-Nghia Le},
  booktitle={ACM Multimedia 2025 Demo and Video Track},
  year={2025},
  url={https://openreview.net/forum?id=lC86otupSN}
}
You can add a tracker to track page visits by creating an account at statcounter.com