--- title: "Deploy OpenShorts" description: "AI short generator with Apify YouTube downloading" category: "AI/ML" url: https://railway.com/deploy/openshorts-1 --- # Deploy OpenShorts AI short generator with Apify YouTube downloading **[Deploy OpenShorts on Railway](https://railway.com/template/openshorts-1)** Machine-readable deploy manifest (JSON, validated by TemplateCI): https://railway.com/deploy/openshorts-1/manifest.json - **Creator:** INAPP - **Category:** AI/ML ## Template content ### openshorts - **Source:** https://github.com/INAPP-Mobile/openshorts-fresh - **Public domain:** Yes ## Documentation # OpenShorts — AI-Powered Vertical Video Generator [![Deploy to Railway](https://railway.app/button.svg)](https://railway.com/deploy/openshorts) OpenShorts turns long-form videos into viral vertical shorts. It offers three tools in one dashboard: 1. **Clip Generator** — paste a YouTube URL, get AI-detected highlight clips cropped to 9:16. 2. **AI Shorts** — generate UGC-style marketing videos with AI actors, subtitles and voice-over. 3. **YouTube Studio** — generate thumbnails, titles and descriptions with AI. This is the official one-click Railway template. The dashboard is served same-origin from the FastAPI backend, so there is no extra service to configure. --- ## Features - **YouTube downloading via Apify** — residential-network downloads bypass YouTube's datacenter-IP bot checks (includes $5/month free tier). - **yt-dlp fallback** — optional local downloader if you prefer not to use Apify (may fail on Railway's AWS/GCP IPs). - **AI clip detection** — Gemini analyses long videos and picks the most viral moments. - **Vertical crop with smart tracking** — auto-faces/peaks keep subjects centred in 9:16. - **Burned subtitles** — faster-whisper transcription, baked into the final MP4. - **AI voice-over and translation** — synthesis + translation pipeline for multilingual shorts. - **AI thumbnail & title generation** — artwork and copy from Gemini. - **S3 backup gallery** — optionally mirror generated clips to AWS S3 for a public gallery. - **Hash-routed dashboard** — single-page React app served by FastAPI (no separate web server). - **One worker-concurrency limit** — stay inside a small Railway plan's RAM envelope. ## Quick Start 1. Click the **Deploy to Railway** button above. 2. Railway will prompt you for `GEMINI_API_KEY` — this is required for the full pipeline. Supply it during setup (or add it in the service's Variables tab afterward). 3. (Optional) Set `APIFY_TOKEN` for YouTube downloading via Apify — residential-network downloads bypass YouTube bot checks. Free tier includes $5/month usage. 4. Wait for provisioning and the build to finish (first build pulls torch + deps; allow 3–5 minutes). 5. Open your service URL — the dashboard loads at `/`. 6. Use the **Clip Generator** tab with a YouTube URL, or upload a local video file. ## YouTube Cookies (Optional) Some YouTube videos are age-restricted or require login. To download these, provide a browser cookies file: 1. Install a cookies export extension (e.g. [Get cookies.txt LOCALLY](https://chrome.google.com/webstore/detail/get-cookiestxt-clean/ahmfnjeglddpplmkeijbbdbbjhofmlkn) for Chrome). 2. Log into YouTube in your browser. 3. Export your cookies to a `cookies.txt` file (Netscape format). 4. In the OpenShorts dashboard, go to **Settings** → **YouTube Cookies** and upload the file. The file is saved to a fixed path on the Railway volume and is picked up by yt-dlp immediately — no service restart or path configuration needed. ## Environment Variables | Variable | Required | Description | |----------|----------|-------------| || `GEMINI_API_KEY` | Yes | Google Gemini API key for AI analysis, titles/thumbnails, voice-over. Get one at https://aistudio.google.com/apikey. Can also be passed per request via the `X-Gemini-Key` header. | || `APIFY_TOKEN` | No | Apify token for YouTube downloading. Residential-network downloads bypass YouTube's datacenter-IP bot checks. Free tier includes $5/month usage. If unset, falls back to yt-dlp. | || `MAX_CONCURRENT_JOBS` | No | Max parallel video jobs. Each is CPU/RAM heavy. Default `2`. | | `AWS_ACCESS_KEY_ID` | No | Optional AWS access key for S3 upload of generated clips. | | `AWS_SECRET_ACCESS_KEY` | No | Optional AWS secret key paired with `AWS_ACCESS_KEY_ID`. | | `AWS_REGION` | No | Optional AWS region. Default `eu-west-3`. | | `AWS_S3_BUCKET` | No | Optional S3 bucket for generated clips. | | `AWS_S3_PUBLIC_BUCKET` | No | Optional public S3 bucket for gallery public URLs. | | `PORT` | No | HTTP port (Railway sets this; default 8000 in the image). | ## Prerequisites - A [Railway account](https://railway.app). - A [Google Gemini](https://aistudio.google.com/apikey) API key. - (Optional) an AWS account + S3 bucket if you want the public gallery. ## Architecture ``` ┌─────────────────────────────────────────────────────────┐ │ Railway Project │ │ │ │ ┌─────────────────────────────────────────────────┐ │ │ │ OpenShorts service (single) │ │ │ │ │ │ │ │ uvicorn ── app:app (port $PORT) │ │ │ │ ├── /api/* JSON API (process, status…) │ │ │ │ ├── /health Railway healthcheck │ │ │ │ ├── /videos/* static generated clips │ │ │ │ ├── /thumbnails/* static thumbnails │ │ │ │ └── /* pre-built React dashboard │ │ │ │ │ │ │ │ Volume: /data ── output/ + uploads/ │ │ │ └─────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────┘ ``` The multi-stage Docker build compiles the Vite dashboard in a `node` stage, then copies the result into a `python:3.11-slim` runtime image with CPU-only PyTorch, ffmpeg and all ML deps. There is no separate renderer to run — the dashboard's preview is in-browser (`@remotion/player`). ## Deploy Click the button at the top of this README. Railway builds the image, mounts a persistent volume for your generated clips, and routes a public domain to the service. If you set a custom domain later, update the service's **Settings → Networking → Custom Domains** page. ## How It Works 1. You submit a YouTube URL (or upload a file) with a Gemini key. 2. The backend queues the job and spawns `main.py` in a subprocess. 3. If a YouTube URL was given, yt-dlp fetches the source MP4 locally (no vendor lock-in). 4. Gemini picks highlight moments; ffmpeg crops each to 9:16 with smart tracking. 5. `faster-whisper` transcribes audio; subtitles are burned in. 6. The finished clips land under `/data/output//` and are streamable at `/videos/...`. --- # Deploy and Host Deploy this template on Railway with one click. Railway provides compute, TLS at the edge, and a public URL. The service restarts automatically on failures. ## About Hosting This template runs as a single container with a persistent volume for generated videos and uploaded source files. The dashboard, the JSON API, the healthcheck and the video files are all served by one FastAPI process on `$PORT`. There is no external database to manage — clips are written directly to the attached volume. ## Why Deploy - **One-click deploy** — no setup, just provide two API keys and go. - **Single service** — no multi-service orchestration, no extra Redis/Postgres to wire. - **Automatic HTTPS** — Railway provisions TLS certificates automatically. - **Self-healing** — automatic restarts on failure. - **Persistent storage** — your generated clips survive deploys and restarts on the attached volume. - **CPU-only ML** — PyTorch runs on CPU, which is all a default Railway plan needs. ## Common Use Cases - Repurpose long-form YouTube/Twitch content into TikToks, Reels and Shorts. - Produce AI-UGC marketing videos at scale. - Generate AI thumbnails and titles for a video pipeline. - Run a self-hosted alternative to paid clipping SaaS. - Prototype a short-video pipeline on a free Railway plan. ## Dependencies for OpenShorts ### Deployment Dependencies - [Railway Account](https://railway.app) — hosting platform. - [yt-dlp](https://github.com/yt-dlp/yt-dlp) — YouTube video downloader (bundled with the image, no external token needed). - [Google Gemini API](https://aistudio.google.com/apikey) — AI clip detection, titles, thumbnails, voice-over. No external database, cache or message queue is required. Optional AWS S3 is only needed if you want generated clips mirrored to a public gallery. ## Similar templates - [Chat Chat](https://railway.com/deploy/-WWW5r) — Chat Chat, your own unified chat and search to AI platform. - [stella](https://railway.com/deploy/stella) — Self-host stella with web, API, Postgres, Redis, and object storage. - [Hermes Agent | OpenClaw Alternative with Dashboard](https://railway.com/deploy/hermes-agent-or-openclaw-alternative-wit) — Self-Hosted Hermes AI Agent for Telegram, Discord & Slack Open this page in a browser: https://railway.com/deploy/openshorts-1