Long2Text

Long screenshot to formatted text. Specialized OCR for chat records, meeting transcripts, and articles.

Architecture

long2text/
├── web/     # Next.js frontend (Vercel)
├── api/     # FastAPI + PaddleOCR backend (Railway/Docker)

Quick Start

Frontend (Next.js)

cd web
npm install
npm run dev     # http://localhost:3000

Backend (FastAPI + PaddleOCR)

cd api
pip install -r requirements.txt
uvicorn main:app --reload   # http://localhost:8000

Docker (Backend)

cd api
docker build -t long2text-api .
docker run -p 8000:8000 long2text-api

Deploy

Frontend → Vercel

cd web
npx vercel --prod

Backend → Railway / Fly.io / Self-hosted

The Python backend requires PaddleOCR which needs ~2GB RAM.

Option 1: Railway

cd api
railway up

Option 2: Docker on your Mac Mini

cd api
docker compose up -d

Environment Variables

See web/.env.example for all configuration options.

Features

• Smart splitting of ultra-long screenshots with overlap
• PaddleOCR for high-accuracy Chinese + English recognition
• Scene-specific formatting (chat, meeting, article)
• Markdown output preserving structure
• Bilingual UI (Chinese/English)
• Free tier + paid full results