R&D in AI for Cost Reduction in Product Photography
Mission
As the Product Designer on this project, my mission was to lead the R&D effort in identifying and testing AI solutions to reduce a $300,000 annual budget spent on product photography. By leveraging generative AI technologies and conducting iterative testing, I aimed to design a framework that could seamlessly integrate AI-generated scenes with the client’s existing workflows. This included maintaining the integrity of the imported products and ensuring realistic, high-quality backgrounds for catalog imagery.
Team structure
1 Product Designer (me)
3 Developers
Research methods
Generative AI exploration (MidJourney, DALL-E, HuggingFace, Mokker.ai)
Competitor analysis of existing AI tools
Iterative testing and validation of generated visuals
Co-creative workshops for framework adoption
My role
Product design, Visual framework creation for prompt generation, Co-hosting workshops and training sessions with client teams
Plateform
Web Application (prototype with iframe API integration)
Time Frame
Complete in 2 months 2024
Design tools
Figma, Mokker.ai, Whimsical
R&D
Product Design
AI
LLM
Prompt
Overview
An international furniture and home decor company approached our team with a specific challenge: to reduce their annual $300,000 budget spent on product photography for their catalog. Their current process involved shipping a container of furniture and decor items to Spain, renting a location and a production crew, and capturing high-quality photos for their magazine distributed to clients worldwide.
Objectives
Deliver a scalable framework to integrate AI-generated visuals into the client’s existing workflows seamlessly.
Constraints
01
The AI-generated images must not alter the imported product’s shape, color, or texture.
Product Integrity
02
The generated scenes (e.g., houses, outdoor settings, beaches) must look authentic and not artificial.
Realistic Backgrounds
Outcomes
The best-performing AI solution, Mokker.ai, achieved only a 12% reduction in production costs, far from the target of 80%.
01
Most AI tools tested distorted product shapes or generated unrealistic backgrounds.
02
The immaturity of current AI technology limited its potential to fully replace traditional photography workflows.
03
Research and Exploration
-Approach-
Our team, comprising myself as a Product Designer and several developers, conducted extensive research into existing AI tools and technologies. We evaluated various platforms to find a solution that met the client’s constraints.
Explored Solutions
01
We tested image generators such as MidJourney, DALL-E, and platforms available on HuggingFace. While these tools offered impressive capabilities, they consistently distorted the imported products and produced backgrounds that appeared artificial.
General AI Platformsimplicity and Functionality
02
Further exploration led us to Mokker.ai, which showed the most promising results. Mokker.ai allowed us to generate realistic backgrounds while maintaining product integrity. This became the cornerstone of our proposed solution.
Specialized AI Tools
Testing and Validation
Testing the Tools
We conducted multiple tests to validate Mokker.ai capabilities
Ensured the generated visuals met the client’s standards.
Image Qualitys
Evaluated the authenticity of generated environments.
Background Realism
Verified that the imported products were not distorted or altered.
Product Accuracy
Client Feedback
Initial results were presented to the client through
Sharing generated samples for their feedback.
Image Reviews
Engaging their team to interact with the tool.
Team Workshops
Suggested Image Placement
Add the "Scene Validation" screenshot to illustrate the variety of generated backgrounds and their quality.
Although the client appreciated the quality, some members faced challenges in adapting to the new workflows. This led us to develop additional support materials.
Framework Development
To simplify adoption, I designed a framework to streamline the use of Mokker.ai within the client’s team
1. Prompt Framework
Created a text-based prompt template with fill-in-the-blank fields for generating visuals based on specific requirements (“Generate a [product] in a [location] with [lighting conditions]”).
2. Visual Prompt Templates
Designed posters and visual guides to help team members visualize and construct prompts effectively.
3. Visual Prompt Templates
Organized a hands-on workshop where participants learned to use Mokker.ai. While some adapted quickly, others required additional support, revealing the need for continuous refinement of the framework.
Prototyping and Application Design
Conceptualizing the Application
Our team began prototyping an application to integrate Mokker.ai’s capabilities seamlessly into the client’s workflow. Key features included:
An iframe within the application to host Mokker.ai’s functionality.
API Integration
A folder system for uploading and organizing product images.
Image Management
Direct API connection to streamline background generation and product placement.
Automated Workflow
Challenges Encountered
Despite the promising prototype, the client’s tests revealed that the tool only achieved a 12% reduction in production costs, far from the 80% target. This gap led the client to conclude that the solution was insufficient for their needs at this stage.
Insights and Learnings
Key Insights
01
Many AI platforms, including Mokker.ai, are not yet advanced enough to handle the nuanced requirements of high-quality product photography.
Immaturity of AI Tools
02
Successful integration of new tools requires intuitive frameworks and comprehensive training to ensure user adaptability.
Adoption Barriers
03
While AI can provide incremental improvements, significant cost savings may require more mature technology or additional process reengineering.
Cost-Benefit Analysis
Learnings
01
This project deepened my understanding of generative AI’s capabilities and limitations, especially in maintaining product integrity.
Understanding AI Limitations
02
Developing tailored frameworks for client-specific use cases is essential for adoption and scalability.
Framework Design
03
Even though the client did not adopt the solution, the R&D process offered valuable insights into AI tools and their potential for future applications.
Value of R&D
Conclusion
This project was an invaluable exploration of AI’s potential to reduce costs in traditional workflows. While the solution fell short of the client’s goals, it paved the way for future innovations and highlighted critical areas for improvement in generative AI tools. The experience also strengthened our team’s expertise in working with emerging technologies and crafting client-focused solutions.
