cat leorodney.ai
11 / 19
leorodney
An AI image generator: type a prompt, get an image. For artists, designers, anyone.
- role
- founder, main developer
- year
- 2023
- access
- -r--r--r-- (archived, read-only)
- problem
- Image generation was new in 2023 and lived behind APIs. I wanted a site where anyone types a prompt and gets an image back.
- approach
- A React and Redux client over Node services: accounts, a community feed of prompts, DALL·E to draw, and GPT-3 to write a prompt when you have none.
- outcome
- Built as microservices, then rewritten as one app six weeks later. It's offline now; the source isn't.
TL;DR
// overview
Architected as cloud-native microservices on the MERN stack and Docker: an auth service, a gateway, prompts and a router behind a React and Redux client.
Prompts go to OpenAI's DALL·E, images are hosted on Cloudinary, and users can download and share what they generate. Authentication and access control run throughout.
// microservices, then a monolith
The first server was five services behind an nginx router, each with its own Dockerfile: auth, a gateway, prompts, production and promptify.
Six weeks later the same routes came back as one Express app.
// surprise me
Out of ideas? GPT-3 writes a prompt, DALL·E draws it at 1024 by 1024, and Cloudinary keeps the result for the feed, ready to download or share.
// the basics, even here
helmet, lusca, rate limiting, bcrypt for passwords and sessions stored in MongoDB, with CodeQL scanning the code. Every service also ran on @handy.js/handy, the library that became handyscript.
// stack
- typescript
- mern
- docker
- dall·e
- cloudinary