ModeMixer

ModeMixer: Mix Trends, Master Style.

Comment 2

ModeMixer: The Ultimate AI Fashion Design Pipeline for Customized, Trend-Driven Styles

Inspiration

The fashion industry has long been a fusion of creativity and practicality. ModeMixer was born from our shared passion for fashion and technology, with the aim of bridging the gap between artistic design and technical execution.

The idea originated when a friend, who owns a small fashion retail business, shared her challenges in keeping up with the latest trends while catering to her customers' diverse tastes. She found it difficult to source unique designs quickly and affordably, often feeling like she was playing catch-up in a fast-moving industry. On top of that, she struggled with the manufacturing process, particularly in generating detailed tech packs that would translate her vision into production-ready specifications.

We wondered, "What if we could harness the power of AI to help small business owners like her create unique fashion designs that are aligned with current trends, while also simplifying the manufacturing process?"

Thus, ModeMixer was envisioned as an end-to-end fashion design pipeline, enabling users to:

  • Collect Trends: Stay updated with the latest fashion trends and incorporate them into their designs.
  • Customize Designs: Provide a name and description to generate unique fashion designs tailored to their vision.
  • Create Tech Packs: Receive a comprehensive tech pack that includes details like fabric, measurements, and construction techniques.

The name "ModeMixer" reflects our vision perfectly:

  • Mode: Associated with fashion (à la mode) and a particular method of doing something.
  • Mixer: Suggests the blending of trends, styles, and data to generate trend-aligned clothing designs.

Overall, ModeMixer emphasizes the dynamic, innovative process of combining existing fashion elements to create fresh, trendsetting designs using AI technology.

What It Does

  1. Explore Trends: ModeMixer scrapes the internet for the latest fashion images and has a database of the latest fashion styles. It incorprates these into a User's design whenever a user creates it new Collection. This allows users to:

    • Incorpate the Latest Fashion Trends into your Brand Identity: Focus on your brand Identity while also being able to stay on top of the latest trends.
    • Gain Inspiration: Browse through curated collections to inspire your own designs.

  2. Generate Design: In this step, users can:

    • Decide the Name of the Collection: Provide a name to represent the design collection.
    • Describe the Collection: Users can either let the web app generate a style description based on the collection name or manually add personal customization and brand identity to the description.
    • Generate Relevant Clothing Items: Create five clothing items based on user input and current fashion trends.

  3. Create Tech Packs: For each clothing item, ModeMixer generates a comprehensive tech pack that includes:

    • Fabric Type and Treatment: Details on the type of fabric used and any treatments applied.
    • Measurements: Specific measurements for different sizes.
    • Graphics, Adornments, and Hardware: Any graphics, adornments, or hardware associated with the item.
    • Size Quantities: Information on size quantities for production.
    • Interior Tags: Details on labels and interior tags required.

How We Built It

  1. Backend Infrastructure Setup:
    • FastAPI: Used as the API framework.
    • MongoDB: Serves as the vector database, acting as both the latest fashion knowledge base and the database for generated collections and items.
    • Databricks Workspace:
      • Web scrapes and uploads fashion images to the AWS S3 bucket.
      • Generates vectors of stored fashion image descriptions.
  2. Web App Workflow:
  • Web Scraping: Scraped thousands of male and female celebrity fashion images and stored them in the AWS S3 bucket using Databricks Workspace.
  • Image Description Generation:
    • Generated descriptions for each image using GPT-4.
    • Vectorized the descriptions using OpenAI embeddings and stored them in the MongoDB vector database to populate our fashion style knowledge base.
  • Collection Description Generation:
    • Used Databricks' DBRX model to generate a collection description based on the collection name.
    • The collection description was fed into a well-designed prompt template to generate five item names.
  • Item Description Generation:
    • For each item, passed the item name and the most relevant celebrity fashion reference (obtained from similarity search in the vector database) as context to generate the item description.
  • Design Image Generation:
    • Passed the item description into the DALLE model to generate the final design image.
  • Tech Pack Generation:
    • Feed the design image back into GPT-4-V to acquire a more detailed description.
    • Passed the detailed description to the Databricks DBRX model to generate all sections of the tech pack in markdown format concurrently for better speed.
  • Tech Pack Conversion to PDF:
    • Converted the tech pack from markdown to PDF for the user to download and send to manufacturers.
  1. Frontend Setup:
  • React.js: Used for building the user interface.
  • Material Tailwind and Tailwind: Set up the interface that allows users to:
    • Enter their customized collection name and description.
    • Create and browse their generated collections.
  • Features:
    • View the all AI-generated collections. In each collection, display design item image, item name, item description, and the celebrity fashion reference that inspired the design.
    • Edit each item after creation
    • Download the tech pack for each item.

Challenges We Ran Into

  1. Handling Large Numbers of API Calls:
    • Implemented concurrency to handle the high volume of API calls required for generating descriptions, images, and tech packs.
    • Utilized asynchronous processing to improve the speed and responsiveness of the application.
  2. Model Optimization:
    • Achieving high-quality and diverse designs required iterative refinement through prompt engineering.
    • Adopted speedy models like Databricks' DBRX to increase the processing speed of generating descriptions and tech packs.

Accomplishments That We're Proud Of

  1. Seamless Integration: Successfully integrated the AI models with the web application for real-time design generation.
  2. Diverse Design Generation: Achieved high-quality, diverse fashion designs using LLMs.
  3. Tech Pack Creation: Developed a comprehensive tech pack generation feature that translates designs into production-ready specifications.
  4. User-Friendly Interface: Created an intuitive and interactive interface for users to easily create and explore designs.

What We Learned

Developing ModeMixer was a journey of exploration and learning. We deepened our understanding of:

  • Fashion Trends and Design: Analyzing current trends and translating them into AI-generated designs.
  • AI and ML Models: Working with various machine learning models, particularly in the field of generative design.
  • Tech Pack Creation: The intricacies of creating comprehensive tech packs for manufacturing.
  • Databricks Ecosystem: Utilizing Databricks' collaborative and scalable environment for machine learning and data processing.

What's Next for ModeMixer

  1. Enhanced Personalization: Allow users to provide more specific inputs for design generation, such as color preferences, patterns, and styles.
  2. Collaborative Design Space: Introduce a feature for collaborative design, allowing multiple users to co-create designs.
  3. Integration with Manufacturers: Partner with fashion manufacturers to provide a smooth transition from design to production.
  4. Expanded Trend Analysis: Incorporate real-time trend analysis using social media and fashion industry reports.
  5. Mobile Application: Develop a mobile version of ModeMixer for on-the-go design creation.

Built With

Share this project:

Updates