Cursor for Image Editing: A Multi-Agent Approach for Visual Content Creation and Editing

In the fast-evolving world of digital advertising, crafting visually stunning and brand-consistent visuals can be a time-consuming and costly endeavor. Our team at PES University set out to tackle this challenge with a novel multi-agent system that redefines how we generate and refine visual content. Our work, recently presented at the Multi-Agent Workshops at AAAI 2025, introduces an innovative approach to creating posters, banners, and flyers with unprecedented control and efficiency.

The Challenge with Current Image Generation Models

Foundational models like Stable Diffusion excel at producing high-quality images from text prompts. However, they often fall short when it comes to post-editing capabilities. Need to tweak the layout, align text perfectly, or swap an object? You're often left resorting to manual editing tools or complex techniques like inpainting. This limitation is especially problematic in industries like advertising, where precision and brand consistency are non-negotiable.

Our Solution: A Multi-Agent System for Iterative Refinement

Our system combines the power of Large Language Models (LLMs) and Vision-Language Models (VLMs) to create a seamless workflow. Here's how it works:

Initial Generation

We start by segmenting key objects from provided images, crafting a narrative that ties them together in a cohesive story, and generating an initial image.

Iterative Refinement

Specialized agents analyze the image for visual inconsistencies, such as misaligned text or clashing colors, and propose targeted edits. This mimics the human editing process while allowing for manual tweaks.

Customization

Users can fine-tune layouts, fonts, and effects through an integrated editor, ensuring the final image meets brand guidelines and marketing objectives.

This multi-agent architecture not only enhances control but also accelerates the creative process, making it ideal for producing consistent, high-quality visuals at scale.

Why It Matters

Our approach addresses the repetitive and resource-intensive nature of traditional design workflows. By automating iterative refinement, we empower designers to focus on creativity rather than tedious adjustments. Whether it's a vibrant poster for a university or a sleek banner for a product launch, our system ensures visual coherence and brand alignment with minimal effort.

Curious to see it in action? Check out our

or dive into the full paper presented at the Multi-Agent Workshops at AAAI 2025.

Acknowledgments

This project was a true team effort. A huge shoutout to my talented teammates— Achala, Nigel, and Srinidhi— and our mentor, Prof. Srinath, for their invaluable contributions and guidance.