Describe your project

Our project harnesses the power of Generative AI (GenAI) to dynamically generate promotional banners and videos based on user inputs such as product images, promotional offers, color palettes, and pre-defined themes. By leveraging the capabilities of Gemini 1.5 Pro on Google Vertex AI, we interpret user input and generate detailed, structured banner descriptions.

The system then transforms these descriptions into four unique and visually appealing banners for each input, offering creative variations in text placement, background design (solid colors, gradients, or image-based), and overlay elements like decorative icons or badges. This ensures that the promotional content is both aesthetically pleasing and aligned with the user's branding requirements.

What makes our solution stand out is its focus on non-designers, allowing them to effortlessly create engaging promotional content that maintains brand consistency while incorporating creative design elements. The flexibility of the system ensures that users can quickly generate high-quality banners without needing design expertise, while still allowing for customization and creative expression.

Using Natural Language Processing (NLP), the system fine-tunes the banner descriptions, ensuring that even with limited resources, we can generate personalized and scalable banner variations. This results in tailored, efficient, and flexible promotional content creation, which adapts to any type of input, making it perfect for businesses looking to streamline their promotional design processes.

Challenges we ran into

One of the primary challenges we faced was the unavailability of advanced image diffusion models, such as Imagen, due to limited access and our student budget. To overcome this, we adapted by integrating external APIs like DALL·E and Hugging Face into our workflow, enabling us to handle text-to-image generation efficiently without compromising on quality.

Another challenge was fine-tuning the Gemini 1.5 Pro model to generate multiple variations of banners from a single set of inputs. Striking the right balance between creativity and efficiency proved difficult, as we had to ensure that each banner variation remained unique while still adhering to brand guidelines. By refining our prompt structures and generating more detailed descriptions, we were able to maintain high precision in the design output while allowing for creative variation.

We also faced scalability and flexibility challenges when processing user-generated product images that came in various formats, orientations, and resolutions. To address this, we introduced flexible input handling and optimized the banner generation process to accommodate diverse image formats. This made the system not only scalable but also user-friendly and adaptable, capable of managing a wide range of input types seamlessly.