Project Status: archived Learn More: donphi.work/works/draftmerch
Draftmerch was a magical tool for creating consistent, high-quality, AI-generated logos, designed specifically for sports fantasy leagues. It was a passion project and a fantastic learning experience, built with ChatGPT-3 acting as a co-writer.
Instead of a simple text prompt, Draftmerch used a unique pipeline. Users would pick from pre-defined lists of characters, sports, fun actions, and environments. This would generate a specific, repeatable prompt and a seeding ID, which was fed into the OpenAI API to render a consistent and fun logo.
Once a design was created, a second serverless pipeline would run:
- Use Python and OpenCV for sophisticated image manipulation and upscaling.
- Send the image to a vectorization API to create a perfect, print-ready vector file.
Ultimately, the project had to be shut down. The operational costs for the AWS services (especially Lambda) and the OpenAI API tokens became too high to sustain for a personal project.
This was a tough decision, as many of my friends' children loved it. A special shout-out to my biggest fan, Eli ❤️, who was heartbroken when it went offline.
The core of this project was the unique prompt and seeding technique used to get consistent results from DALL-E 2. This wasn't a simple, single-line prompt; it was a five-paragraph, hyper-engineered prompt that took months to perfect.
To this day, that prompt remains a secret. However, if you're a fellow developer or researcher, feel free to reach out, and I'll be happy to share it with you.
This project was a full-stack, serverless application.
- Frontend: Hand-coded with love using HTML, CSS, and vanilla JavaScript.
- Backend Logic: Python & AWS Lambda (a 12-stage pipeline).
- AI & Image Processing: OpenAI & OpenCV.
- Database: Amazon DynamoDB.
- API & Real-Time: AWS API Gateway & Websockets.
- CI/CD Pipeline: A seamless CI/CD (Continuous Integration/Continuous Deployment) pipeline using AWS CodeBuild & AWS CodePipeline. This provided automated, seamless updates from GitHub directly to the 12-stage AWS Lambda servers.
- Security: AWS Secret Manager.
As this project is archived, it is no longer maintained. Running it would be a complex undertaking, requiring:
- Cloning the repository.
- Setting up the frontend (e.g., serving the static HTML/JS/CSS files).
- Configuring the entire AWS backend, including:
- Deploying the Python code to AWS Lambda.
- Setting up the DynamoDB tables.
- Configuring API Gateway and Websockets.
- Populating AWS Secret Manager with your OpenAI API keys and other credentials.
- Rebuilding the CodePipeline and CodeBuild CI/CD process.