Aurai

A self-hosted web app that brings old photos back to life using AI. It fixes faded and yellowed colors, colorizes black-and-white pictures, and turns a photo into a short video. It can also change a video with a simple sentence.

This is a small project I wanted to create to use Openrouter image and video AI models in a small contained environment. Feel free to create a PR and add more features.

What it does

• Enhance — faded, yellowed or too-red photos get their natural colors back. The people in the photo stay exactly as they are: Aurai keeps your own pixels and only takes the colors from the AI, so no face is redrawn.
• Colorize — natural colors for black-and-white photos, with every face and shadow as it was.
• Repair — removes scratches, dust, creases and small tears. Only the damaged spots change; the rest of the photo stays your own.
• Upscale — makes a small photo bigger and sharper, up to 4K, or a blurry one sharper at its own size. The shapes and colors stay your photo's own, and the AI only adds finer detail. You can also just resize it, for free.
• Edit Image — change a photo with a sentence: "make it a winter day". Paint over a part of it to change only that part, like "a red cap" on his head; the rest of the photo stays your own.
• Animate from Image — a slow camera move over a still photo, where nobody moves. Try several AI models side by side, or make a free Ken Burns zoom on your own server.
• Photos to Video — a few photos taken seconds apart become one short video: the AI fills in the movement between each two of them.
• Edit Video — upload a short clip and write what to change: "make it a snowy winter day".
• Enhance Video — better color and light for dark, flat or tinted videos. The AI corrects a frame every few seconds of each scene, and every frame gets that correction, so nothing in the video is redrawn.
• Before/after slider, small fixes like "a little warmer skin", downloads in full size.
• Send a result on to the next tool: repair a photo, colorize it, edit it, upscale it, then animate it.
• Drop up to 10 photos at once and start them all with one model.
• Mark finished photos and videos as done, to keep the history tidy.
• A history of everything you made. You see the price of a run before you start it, whenever OpenRouter publishes one.
• Sign in with a code sent by email. No passwords.

The AI work is done by models on OpenRouter. Each person connects their own OpenRouter account in Settings and pays OpenRouter directly. The Ken Burns videos are made on your server and are free.

Pick a camera move and a model — or several, to see them side by side:

Running it

You need Docker and an .env file:

cp .env.example .env

Open .env and fill in two secrets, AUTH_SECRET and ENCRYPTION_KEY. They must be different. This makes a good one:

openssl rand -base64 32

Set SITE_URL to the address you will open Aurai at, for example https://photos.example.com.

Then:

mkdir -p data
docker compose up -d

It listens on port 3000. Open it, sign in with your email, and connect OpenRouter on the Settings page. That's it.

The data folder must be writable by the container, which runs as user 1000. If the log says it cannot open the database, run:

sudo chown -R 1000:1000 data

Without compose

docker run -d --name aurai -p 3000:3000 \
  -v "$PWD/data:/app/data" \
  --env-file .env \
  ghcr.io/meceware/aurai:latest

Sign-in emails

Without an email server, the sign-in code is printed to the container log. For one or two people that is enough:

docker logs aurai

To send real emails, fill in the EMAIL_SERVER_* and EMAIL_FROM lines in .env.

Settings in .env

Setting	Needed	What it is for
AUTH_SECRET	yes	Any long random text. Changing it signs everyone out.
ENCRYPTION_KEY	yes	Protects the OpenRouter keys saved in the database. Keep it safe; without it, saved keys cannot be read.
SITE_URL	yes	The address you open Aurai at, no / at the end.
EMAIL_SERVER_HOST, EMAIL_SERVER_PORT, EMAIL_SERVER_USER, EMAIL_SERVER_PASSWORD	no	Your SMTP server, for sign-in emails.
EMAIL_FROM	no	The address emails come from, like [email protected]
SIGNUP_ENABLED	no	false stops new accounts. Your own account keeps working.
ALLOWED_EMAIL_DOMAINS	no	Only these email domains can sign up, for example example.com.
MAX_USERS	no	The most accounts there can be. 0 means no limit.
USER_QUOTA_MB	no	Storage for each person, 2048 MB by default.
TURNSTILE_SITE_KEY, TURNSTILE_SECRET_KEY	no	Cloudflare Turnstile on the sign-in page, against bots.

Putting it online

Aurai is one container and needs nothing else in front of it. Point your tunnel or reverse proxy at port 3000 — Cloudflare Tunnel, Nginx Proxy Manager, Caddy, whatever you already use — and set SITE_URL to the public https:// address.

An HTTPS address matters for two things:

• Edit Video works only with it. OpenRouter does not take a video inside the request, so it downloads it from Aurai, through a private link that works only while that edit is being made.
• Videos finish sooner on screen. OpenRouter tells Aurai the moment a video is ready, instead of Aurai checking every now and then.

Anyone who can open the sign-in page can make an account. After you make yours, you may want SIGNUP_ENABLED=false, or ALLOWED_EMAIL_DOMAINS, or Cloudflare Access in front of it.

Uploads are limited to 30 MB for photos and 90 MB for videos. That is under the 100 MB that Cloudflare lets through on its free plan.

Backups

Everything lives in the data folder: the database, and the photos and videos as normal files. To take a backup while Aurai is running:

docker exec aurai node scripts/backup.js

This writes a safe copy of the database to data/backups (the last 14 are kept). Then copy the whole data folder somewhere else. A backup is only a backup once you have tried to restore it.

Accounts

docker exec aurai node scripts/admin.js users
docker exec aurai node scripts/admin.js delete [email protected] --yes

Deleting an account removes all of its photos and videos too. People can also delete their own account in Settings.

Updating

docker compose pull
docker compose up -d

The database is updated on start. Images for amd64 and arm64 are published to ghcr.io on every push to main and on every v* tag.

Building the image yourself

docker build -t aurai .
docker run -d --name aurai -p 3000:3000 -v "$PWD/data:/app/data" --env-file .env aurai

Development

You need Node 24 and ffmpeg.

npm ci
cp .env.example .env.local
npm run dev

Sign-in codes go to the terminal, and to data/dev-outbox.log. To try everything without paying for AI runs, set OPENROUTER_MOCK=true in .env.local: the AI results are then fake ones.

npm test runs the tests, npm run lint checks the code.

Built with

• Next.js
• SQLite via better-sqlite3
• sharp for photos and FFmpeg for videos
• Tailwind CSS + shadcn/ui, with Lucide icons
• Better Auth
• OpenRouter for the AI models

The photo in the screenshots is Guide at Little Norway, Blue Mounds, Wis. by Arthur Rothstein (1942), from the Library of Congress. It is in the public domain; I faded it on purpose to show the restoration.