Degree

Bachelor of Science (Computer Science)

Department

Department of Computer Science

School

School of Mathematics and Computer Science (SMCS)

Advisor

Dr. Ali Raza, Assistant Professor, Department of Computer Science

Keywords

GenAI, Optimization, Image Analysis, Image Generation, Advertising, Computer Vision

Abstract

In the social media advertising industry, small business owners and graphic designers typically struggle with limited resources and a lack of data-driven insights into current marketing trends and competitor advertisements. Additionally, the non-financially inclined may not Traditional digital advertising tools require time, effort, platform-specific knowledge and clear creative direction, creating a barrier for entry. We aim to solve this problem with AdVisory, a comprehensive AI-powered tool for research, creation and optimization– streamlining our users’ competitor research, automating their ad creation and optimizing their budget allocation, making social media marketing accessible to all. This problem was presented to us by SocialChamp, a social media management company that required a tool to automate research, planning and design process for small and mid-sized businesses. AdVisory provides a solution through a clean integration of GenAI tools and optimization techniques into a user-friendly tool that allows a user with no prior domain-specific knowledge to research, plan and launch their social media advertisement campaign. It does this through a series of independent modules, beginning with an advertisement generation module, allowing users to prompt models such as GPT and Flux to automatically generate their ad creative, providing options for the user to both integrate their own product images and edit the final ad. The module also automatically conducts competitor analysis, using data in the public domain, as a benchmark for the generated ads. The optimization module determines the ideal budget split for the user, automatically assigning higher budget to the well-performing ads in order to maximize conversions. In addition to this, the insights module analyses images to infer customer personas from the generated ads, to help users understand their target audience more effectively, and improve their marketing strategy.

Tools and Technologies Used

The hardware requirements for the client side device are as follows:

The device should have minimum 4GB RAM and a modern web browser.

The software requirements for the backend server are as follows:

  • Operating System: Ubuntu 22.04 (for backend services)

  • Frontend: Angular, TypeScript

  • Backend: Node.js, Express.js

  • Database: MongoDB

AI Frameworks: TensorFlow, PyTorch

Methodology

The budget allocation module uses a convex optimization approach, modelling the returns with a log function and maximizing over KPIs such as conversion rate.

The competitor analysis module uses data scraped from Meta Ad Library to show the user's competitor's active ads and generate ads inspired by them with our ad and caption generation modules which use FLUX and GPT Image 1. The methodology involves integrating pretrained AI models via APIs to generate or modify visuals and captions based on user prompts and image inputs

The audience prediction model analyzes uploaded images using vision models like CLIP for visual understanding and BLIP for generating descriptive captions. It goes beyond basic image recognition by identifying the image's current visual and conceptual trends. These trends and the image's content are then used to predict which demographic groups (age, gender, income, profession, interests) and visual styles the image is most likely to resonate with. By incorporating trend analysis into its predictions, the model aims to provide more nuanced and timely insights into an image's potential target audience for advertising or content creation purposes

The Product Ad Module enables users to create advertisements without altering the original product image. Users upload a product photo, which is processed using Python-based libraries to remove the background. The image is then enhanced using the Real-ESRGAN model for better quality.Users can either generate a custom background using Stable Diffusion their own prompts or choose from ready-made backgrounds, which also support prompt-based customization. Once a background is selected, users can place the product image onto it, rotate it, adjust its dimensions, and position it freely. They can also set custom canvas sizes tailored for social media platforms, ensuring the final ad is optimized for online posting.

Document Type

Restricted Access

Submission Type

BSCS Final Year Project

Download

Share

COinS