QuickContent.io

About the project

A content autopilot for creators and small businesses. Give it your website, and it learns your brand and makes posts that look and sound like you.

QuickContent.io is a content autopilot for creators and small businesses. You give it your website. It learns your brand, its colours, tone and audience, then makes posts, images and short videos that look and sound like you.

I built it with Jay Prakash Pathak and Krishna. My part was its brain: brand extraction, content and image generation, choosing the right model for each image, video generation and the media library. I also split two npm packages out of it.

How it works

1. 1

   Read the brand

   A background worker crawls your site, stores the pages for search, and writes a brand summary with logo, colours, audience and tone.

2. 2

   Write

   Ask for a post, and it's written in your tone with facts from your own site.

3. 3

   Design

   Each image prompt is sorted first: text-heavy designs go to one model, visual ones to another. Your logo and reference images can be added.

4. 4

   Make video

   Pick one of six video models to turn an idea into a short clip.

5. 5

   Keep

   Everything lands in your media library, in our own storage, ready to post.

Under the hood

Apps

A Next.js web app and an Express API in a Bun monorepo.

Jobs

BullMQ on Redis for crawling and generation, with progress stages.

AI

OpenAI for text and sorting prompts, FAL for images and video.

Brand knowledge

Pinecone or ChromaDB through my kb-client package, crawling through my web-extractor package.

Storage

Postgres with Drizzle, and images in AWS S3 behind signed links.

What I did

• Brand extraction: a background worker crawls the site, stores it for search, and writes a brand summary (logo, colours, audience, tone) with live progress
• Workspaces, onboarding and settings
• AI posts with images, a visual post creator with previews and the right size for each platform
• Smart model choice: each prompt is classified, text-heavy images go to one model and visual ones to another
• Video generation with a choice of six models, saved to the media library
• Brand knowledge in every image prompt, plus logo, reference images and content history
• Generated images saved to our own storage, not temporary links
• Published two npm packages from it: a site crawler (@anisirji/web-extractor) and a knowledge-base client (@anisirji/kb-client), about 1,100 downloads in a year

What was new

• Routing each prompt to the image model that suits it
• Turning internal code into public npm packages

Problems I hit, and how I fixed them

1. 01

   The AI made up logo and image URLs during brand extraction.

   Fix Strict rules, then keep only URLs that were really found.

2. 02

   Regenerating an image overwrote the post text.

   Fix Text is written once, with a separate Regenerate Content button.

3. 03

   Image links expired and private logos didn't load.

   Fix Copy images to our storage and use signed links.

4. 04

   Posts could stay "generating" forever.

   Fix Clean them up after 10 minutes, and charge credits only on success.

Screenshots