Edit photos by describing the change, on your own GPU. A free, open-source alternative to Nano Banana and ChatGPT image editing: photos stay on your computer, no daily limit. One installer for Windows, Mac and Linux sets up the best open model (Qwen-Image 2.1). MCP server for AI agents.
Kokon in Docker (NVIDIA)
The same Kokon as the Linux packages, without a window: you use it from a browser on any device. For NVIDIA graphics cards (CUDA). An image for AMD and Intel cards (Vulkan) will follow once it can be tested.
What you need
- Linux with an NVIDIA card and the proprietary driver (
nvidia-smiworks). Windows works too through Docker Desktop with WSL 2. - Docker with the Compose plugin, and the NVIDIA Container Toolkit.
- About 15 GB of disk for the engine and the model.
Start it
Each release publishes the image as ghcr.io/gabrielelozupone/kokon, so you need only the
compose file, no clone and no build:
mkdir kokon && cd kokon
curl -fsSLO https://raw.githubusercontent.com/GabrieleLozupone/kokon/master/docker/compose.yaml
docker compose up -d
docker exec -it kokon kokon-service --set-password # once; at least 8 characters
Open http://<this computer's address>:39152 in a browser, sign in, and follow the setup: it
downloads the engine and the model (about 12 GB) into the container's data volume, once.
Where things are
| Volume | Inside the container | What |
|---|---|---|
kokon-data |
/data |
engine, models, gallery database, logs |
kokon-config |
/config |
settings |
kokon-gallery |
/gallery |
the images saved to the gallery |
To keep the saved images in a folder of your own, replace the kokon-gallery line in
compose.yaml with a folder Kokon can write to (it runs as user id 1000):
mkdir -p ~/kokon-gallery && sudo chown 1000:1000 ~/kokon-gallery
# - ~/kokon-gallery:/gallery
Everyday use
docker logs -f kokon # what the service is doing
docker exec -it kokon kokon-service --set-password # a new password (every browser signs in again)
docker compose down # stop (the volumes stay)
docker compose pull && docker compose up -d # update to a new version
The latest tag follows every release. To get only the fixes of one version, write for example
ghcr.io/gabrielelozupone/kokon:0.11 in compose.yaml (it stays on 0.11.x); each release also
has its exact tag, such as :0.11.0.
Build the image yourself
From a clone of this repository (takes a while: it compiles the UI and kokon-service):
docker build -f docker/Dockerfile -t ghcr.io/gabrielelozupone/kokon:latest .
docker compose -f docker/compose.yaml up -d
Security
- Nothing is reachable until a password is set. The password is stored as a hash, and 5 wrong tries in a row make that device wait.
- The port is published on every network interface. Docker's published ports bypass host firewalls
such as ufw; to use Kokon only from the computer itself, change the port line to
"127.0.0.1:39152:39152". - The connection is plain HTTP: use it on your home network, and through Tailscale or another VPN from anywhere else. Never forward the port on your router. See remote-access.md.
- Kokon's local API with its token (used by scripts and the MCP server) stays inside the container.