MIT licence · Docker · Railway

Self-hosted AI image generation, with the honest caveat up front

Keou Studio is an MIT-licensed AI product image and video studio you can run on your own servers with Docker or Railway. Two ways to generate, both yours: through your own KIE.AI or Fal.ai API keys — or fully local, with images and upscaling running on your own ComfyUI, no cloud involved.

Get the source on GitHub Try it hosted first

Identical feature set to the hosted studio — no watermark, no public gallery, your brand.

What you host, and what you don't

You host the whole studio: the frontend, the Express API, the PostgreSQL database, the poller and your object storage. For the inference itself you have two options. Cloud: generation calls go out to KIE.AI or Fal.ai on a key you supply — no GPU on your server, no model weights downloaded. Local, since v1.0: point Keou at your own ComfyUI instance (LOCAL_ENGINE_URL) and images — and video, with the Wan 2.2 or LTX-Video models installed — run entirely on your hardware, no cloud involved. Voice and sound effects remain cloud-only — we would rather you know before you clone the repository.

What self-hosting buys you instead: no watermark on anything, nothing published to any gallery, full signed downloads and export-pack ZIPs, your own brand across the interface, and no shared queue — your capacity is whatever your provider key and your server allow. Users can each paste their own key; it stays in their browser and is never written to your server's disk, database or logs.

What you need

Docker in two commands

docker build -t keou . docker run -p 3401:3401 --env-file .env keou # open http://localhost:3401

The image runs as a non-root user and ships a healthcheck on /health. On Railway: deploy from GitHub, add a PostgreSQL service (the connection string is injected for you), set JWT_SECRET, the admin pair and the R2 variables — railway.json and the healthcheck are already in the repository, and auto-deploy on main keeps you on the latest build. The full environment reference is in the documentation.

An open-source alternative to Photoroom-style tools

If you are comparing against hosted product-photo tools like Photoroom, here is the honest framing. Keou covers the product-photography workflow — commercial scenes generated around a pixel-locked product, retouching, format adaptation, export packs — plus product video, voice-overs and sound, organised by client and campaign with review links. What you gain by self-hosting is control: your infrastructure, your data, your costs at provider rates, and an MIT licence with no seat pricing. What you give up is the convenience of a managed service — you operate the server, the database and the storage yourself.

MIT means MIT. Run it for clients, charge them, keep the money, rebrand it. No licence fee, no revenue share, no attribution requirement. The build carries a removable provenance mark naming Kanaky Tech as author — documented openly in the docs, it phones nothing home and gates nothing.

Frequently asked questions

Do the AI models run locally when I self-host?

Yes: point Keou at your own ComfyUI instance (LOCAL_ENGINE_URL) and image generation, polish, remix, adapt and upscaling run entirely on your hardware — and video too, if your ComfyUI has the Wan 2.2 or LTX-Video models installed (auto-detected — without them you get a clear message listing exactly what to add). No cloud, no per-generation cost, nothing leaves your machine. Voice and sound effects still go through KIE.AI or Fal.ai on your own keys — ComfyUI core has no TTS, and we don't ship promises built on custom nodes.

What do I need to run it?

Node.js 20+, PostgreSQL, a Cloudflare R2 bucket (free tier, required for uploads) and a provider key. Dockerfile and Railway config ship in the repo; the app listens on port 3401 and migrates its own schema at boot.

Can I use it commercially?

Yes. MIT licence: client work, paid work, rebranding — all allowed, with no fee, revenue share or attribution requirement.

Clone it, boot it, own it

The same studio that runs at studio.kanaky.xyz — on your servers, under your brand, with none of the hosted-demo limits.