Describe your project
Storyblocks uses an agent-driven system with human oversight to create high-quality, customizable promotional ads for products. The AI iteratively refines visuals, mimicking human designers.
What is in-scope in the solution?
- AI-Generated Visuals: Automated creation of high-quality ads using provided product images, promotional details, and themes (e.g., seasonal or event-based like Diwali or Christmas).
- Agentic-Driven Refinement: The system generates an initial design, which is refined iteratively using an agentic loop. Each iteration enhances the layout, text placement, color coordination, and visual elements, ensuring a professional and cohesive final output.
- Human-in-the-Loop Customization: Users remain in control throughout the process. After AI generates the initial visuals, users can adjust text, images, colors, and other elements, ensuring that the final output aligns with brand guidelines and expectations.
- Brand Consistency: The platform integrates brand-specific color palettes, fonts, and style preferences into the design process, maintaining consistency across all promotional assets.
- Multi-Format Output: Generate content optimized for various formats (e.g., banners, social media posts, videos) based on specific size, resolution, and platform requirements (Instagram, YouTube, web banners).
What is out of scope?
- Full automation of ad creation without human review.
- Generation of highly complex custom animations beyond basic video creation.
What are the future opportunities for this solution?
- 3D Content Generation: Expanding the platform to generate 3D promotional content, allowing businesses to create immersive and interactive ads for AR/VR and next-gen platforms.
- AI-Driven Personalization: Integrating consumer data to create hyper-personalized ads tailored to individual preferences, boosting engagement and conversion rates.
- Omni-Channel Campaigns: Expanding capabilities to support cross-platform
Challenges we ran into
Challenges encountered while building Storyblocks:
Integrating multiple models into a single workflow was difficult, requiring seamless coordination between different AI components.
Had to design a new architecture to address the following issues:
- Hallucinations and inconsistent image output: The models sometimes generated images that didn't match the intended theme or purpose.
- Post-generation image editing limitations: Once images were generated, making edits was difficult and often restricted.
- Multiple failure points: Combining various models introduced potential failures that made the workflow less reliable.
- Accurate text placement: Positioning text properly in images was challenging, often leading to misalignment or poor design.
- Layout consistency: Maintaining a visually appealing and consistent layout across iterations was a persistent issue.
Keeping context between iterations of the refinement loop was essential to ensure coherent results across changes.
Finally, computing resource limitations made it hard to manage the iterative process, especially given the high demand for processing power.