PixelPatchwork

Project Description

PixelPatchwork is a collaborative digital art platform that enables users to iteratively refine AI-generated images through masking, prompting, and a voting system to collectively curate high-quality images. The platform encourages collaboration by allowing participants to build on the work of others, fostering a sense of shared ownership and continuous improvement.

Purpose

The purpose of this project is to:

• Investigate the limitations and possibilities of AI-generated art.
• Study worker skills in writing prompts that lead to high-quality images.
• Harness collaborative creativity through iterative and community-driven processes.

Features

• Image Generation: Workers interact with a custom canvas to mask areas of an AI-generated image and provide prompts for modification.
• Voting System: Workers upvote or downvote images to surface the most creative or high-quality outputs.
• Iterative Workflow: The highest-voted image becomes the seed for the next day, creating a feedback loop to refine and build upon previous results.
• Crowd Contributions: Combines individual creativity with collective evaluation to produce diverse and evolving interpretations of seed images.

---

Setup Instructions

Prerequisites

• Python 3.9 or above
• Flask
• MySQL database credentials
• AWS S3 bucket for image storage
• OpenAI API key for DALL-E

Installation Steps

1. Clone the repository:

   git clone [email protected]:dankim444/pixelpatchwork.git
   cd PixelPatchwork

2. Create a python virtual environment:

   python3 -m venv venv
   source venv/bin/activate   # For Linux/MacOS
   venv\Scripts\activate      # For Windows

3. Install required packages:

   pip install -r requirements.txt

4. Configure .env with AWS, OpenAI, and MySQL credentials
5. Run the app:

   python3 src/app.py
   

System Overview

1. Frontend: Built using HTML5, CSS, and JavaScript, with a custom canvas for user interaction.

   • Landing Page: Gives users detailed instructions to participate in the experiment.
   • Image Generation Page: Allows users to mask areas of the seed image and provide a prompt for targeted modifications.
   • Voting Page: Displays all submitted images for a day, enabling upvote/downvote functionality.

2. Backend: Flask application handles routing, image processing, and create/read/update operations.

   • Integrates with OpenAI's DALL-E API for image generation and storage.
   • Stores and handles logic for user contributions/metadata, voting data, and seed images in a MySQL database.

3. Database: MySQL database with three main tables:

   • User: Tracks user details.
   • Image: Stores image metadata, number of upvotes/downvotes, and prompts.
   • Day: Tracks daily seed images, winner of the day, and participation statistics.

4. Quality Control:

   • Voting Mechanism: Users upvote or downvote images to collectively surface high-quality submissions. The voting stage allows us to evaluate the community's perception of quality.
   • Prioritized Display: Past winners are prominently displayed to inspire creativity, highlight exemplary work, and set benchmarks for future contributions.

5. Aggregation:

   • Past Winners: We aggregated a list of past winning images so users could see
   • Comparison of Aggregated vs. Individual Responses: We compared aggregated results with individual vote patterns to understand how collective feedback differed from personal preferences. This comparison highlighted the value of aggregation in surfacing the most broadly appreciated contributions.
   • Vote Distribution: We investigated the distribution of upvotes and downvotes across all images to identify patterns in user engagement

6. Storage: AWS S3 bucket for storing images.

Flow Diagram

Future Work

• Expand the platform for multilingual support to engage a more diverse audience.
• Enhance scalability through database optimization and distributed architecture.
• Implement gamification elements to encourage higher engagement and quality submissions.