Quickstart#

You need a Linux x86_64 server with Docker, 4+ CPU cores (8+ recommended), 8 GB RAM (16 GB recommended) and 5 GB free disk. See Configuration and sizing.

We sent you, separately:

  • the registry address (<registry>, an Amazon ECR host in eu-central-1) and your repository wbg-pro-selfhost/<your id>
  • the current version tag, e.g. 10.12.4-1a2b3c4
  • pull credentials (an access key pair), unless we granted your own AWS account access

1. Log in to the registry#

With the AWS CLI (v2):

Shell

export AWS_ACCESS_KEY_ID=...        # from us
export AWS_SECRET_ACCESS_KEY=...    # from us
export AWS_DEFAULT_REGION=eu-central-1
REGISTRY=...                        # from us, e.g. 123456789012.dkr.ecr.eu-central-1.amazonaws.com
aws ecr get-login-password | docker login --username AWS --password-stdin "$REGISTRY"

The login lasts 12 hours, which is plenty for a pull. If we granted your AWS account instead, run the same docker login line with your own AWS credentials.

Shell

IMAGE="$REGISTRY/wbg-pro-selfhost/<your id>:<version>"
docker pull "$IMAGE"
# Optional, so your servers never depend on our registry:
docker tag "$IMAGE" registry.example.internal/withoutbg-pro-selfhost:<version>
docker push registry.example.internal/withoutbg-pro-selfhost:<version>

The image is yours to keep and run (the license is perpetual). Pull access to new versions lasts for the 12 months of updates in your Order Form.

3. Run it#

Shell

docker run -d --name withoutbg --restart unless-stopped -p 8000:8000 "$IMAGE"

or, with the kit's docker-compose.yml (health check and settings included):

Shell

WITHOUTBG_IMAGE="$IMAGE" docker compose up -d

The model loads at start (up to a minute). Wait until it is ready:

Shell

curl -s localhost:8000/ready     # {"status":"ready"}  (503 while loading)

The image is x86_64 (linux/amd64). On an Apple Silicon Mac it runs under emulation for development (add --platform linux/amd64), much slower than on a server.

4. Check it#

From the kit folder:

Shell

examples/smoke_test.sh                  # or: examples/smoke_test.sh http://<host>:8000

It sends the sample photo through every endpoint, including a Smart Brush edit, and prints All N checks passed. If a check fails, see Troubleshooting.

5. Remove a background#

Shell

curl -s -H 'Content-Type: image/jpeg' --data-binary @photo.jpg \
  localhost:8000/v1/remove-background -o cutout.png

cutout.png is the photo with a transparent background. For the alpha matte only, add ?output=matte. Fixing a cutout with Keep / Remove strokes, JSON requests and all other options: API reference. Interactive docs: http://localhost:8000/docs. Python, Node.js and curl clients: Examples. With an AI coding assistant, give it AGENTS.md (the server also serves it at /llms.txt).

The service listens on plain HTTP and has no authentication: keep it on a private network or put your own reverse proxy (TLS, auth, body size limit) in front of it.

This page as markdown: /pro-model-self-host/docs/quickstart.md. Not a customer yet? Pro Model Self-Host.