Image Gen Model Train - Upload
Landing Page
Image Gen Model Configure
Generate Images
Data Labelling
Phantom Wallet-1
Phantom Wallet-2
Data Labelling Image-Upload
Example Image Gen: 3
Example Image Gen: 1
Example Image Gen: 2

SynthLabel

Inspiration

SynthLabel appears to be inspired by the growing need for effective data labeling solutions in AI and machine learning workflows. The project combines AI image generation capabilities with blockchain technology, likely aiming to create a decentralized platform where users can both generate synthetic data and participate in data labeling tasks. The inspiration seems to come from addressing challenges in obtaining properly labeled datasets for AI training while leveraging blockchain for transparency and incentivization.

What It Does

SynthLabel is a dual-purpose platform that offers:

AI Image Generation: Users can create custom AI-generated images through a web interface, with options to train custom models by uploading their own reference images (up to 30 images per session).
Data Labeling Platform: The system provides tools for uploading images for labeling tasks, with a blockchain-based payment system integrated to reward contributors. The second part still needs to be implemented :(

The platform connects these capabilities through a clean, modern UI built with Next.js and leverages the Solana blockchain for transactions. It uses FAL.ai's services for AI model training and image generation, with workflow steps that include uploading images, training custom models, and generating new images based on these models.

How We Built It

SynthLabel is built using a modern tech stack structured as a monorepo:

Frontend:

Next.js 15 with React 19
TypeScript for type safety
TailwindCSS for responsive styling with custom animations
Solana wallet adapters for blockchain integration
Radix UI for accessible component primitives

Backend:

Bun as the JavaScript runtime
Express.js for the API server
Prisma ORM for database operations
PostgreSQL (via Neon DB) for data storage
AWS S3 for file storage and image hosting
FAL.ai integration for AI model training and image generation

Infrastructure:

Docker for containerization
Turbo for monorepo build optimization

The application follows a microservices architecture with separate frontend and backend services that communicate via RESTful APIs. The system implements webhooks to track the status of AI model training processes and update the database accordingly.

Challenges We Ran Into

Integration with FAL.ai: Implementing and debugging the AI model training pipeline, as evidenced by the complex webhook handling in the backend.
Blockchain Integration: Setting up the Solana wallet connection and payment system, including transaction confirmation.
Asynchronous Processes: Managing the state of long-running processes like model training and incorporating proper error handling.
S3 Integration: Setting up secure file uploads and retrievals while managing permissions.
Database Design: Creating an efficient schema that supports both the image generation and data labeling aspects of the platform.

And a lot's of errors

Accomplishments I am Proud Of

End-to-End Implementation: Successfully integrating multiple complex technologies (AI, blockchain, cloud storage) into a cohesive product.
Modern Architecture: Creating a well-structured monorepo with clear separation of concerns.
Blockchain Payments: Proof of concept, implementing Solana-based payments for data labeling tasks.
Sophisticated UI: Developing a visually appealing interface with animations and responsive design.
Custom AI Model Training: Enabling users to train their own AI models for image generation.

What I Learned

The development process likely provided learning opportunities in:

AI Model APIs: Working with cutting-edge AI image generation models through FAL.ai.
Blockchain Development: Implementing practical use cases for blockchain technology beyond speculative applications.
Webhook Architecture: Creating robust systems for handling asynchronous processes and callbacks.
Monorepo Management: Setting up and maintaining a complex project structure with multiple interconnected services.
Cloud Infrastructure: Managing various cloud services (AWS S3, Neon DB) in a coordinated manner.

What's Next for SynthLabel

Potential next steps for SynthLabel could include:

Rate Limiting : I blocked training image feature, as it is too expensive, will add a payment or rate limiter in future
Enhanced Model Customization: Adding more parameters and options for fine-tuning AI models.
Quality Assurance Tools: Implementing validation mechanisms to ensure the quality of labeled data.
Complete Web3 Part: Some part's for worker side and overall blockchain part needs to be developed for future releases.
Expanded Blockchain Features: Further leveraging Solana for decentralized governance or token-based incentives.
Mobile Application: Developing a companion mobile app for on-the-go data labeling.
Community Features: Adding collaboration tools to enable teams to work together on labeling projects.
Market Expansion: Creating specialized tools for different industries (medical imaging, autonomous vehicles, etc.).
API Access: Developing a public API to allow integration with other AI and ML platforms.