Short answer: yes—you can set a different container image per deployment with

job_variables

in Prefect 3.x. The key is to scope the image value per deployment. If you template a single value at the top (or use one build step output for all), every deployment will get the same image. Here are working patterns you can use from CI: Pattern A: Per-deployment env var interpolation - Put the image directly in each deployment’s

work_pool.job_variables.image

, and populate different env vars for each deployment from CI. prefect.yaml

deployments:
  - name: flow-a
    entrypoint: flows/flow_a.py:flow
    work_pool:
      name: my-k8s
      job_variables:
        image: "{{ $FLOW_A_IMAGE }}"
        namespace: "prefect"

  - name: flow-b
    entrypoint: flows/flow_b.py:flow
    work_pool:
      name: my-k8s
      job_variables:
        image: "{{ $FLOW_B_IMAGE }}"
        namespace: "prefect"

GitHub Actions (two steps, one env var per deploy)

- name: Deploy flow-a
  run: |
    export FLOW_A_IMAGE=<http://ghcr.io/org/flow-a:${GITHUB_SHA}|ghcr.io/org/flow-a:${GITHUB_SHA}>
    prefect deploy --name flow-a

- name: Deploy flow-b
  run: |
    export FLOW_B_IMAGE=<http://ghcr.io/org/flow-b:${GITHUB_SHA}|ghcr.io/org/flow-b:${GITHUB_SHA}>
    prefect deploy --name flow-b

Tip: If you run a single

prefect deploy --all

, any env vars you set will be shared across deployments—so use separate env vars (and/or deploy each by name) when you want different images. Pattern B: Deployment-scoped image build steps - Define a build step per deployment and reference its output in that deployment’s

job_variables.image

. This lets you run

prefect deploy --all

and still get different images. prefect.yaml

deployments:
  - name: flow-a
    entrypoint: flows/flow_a.py:flow
    build:
      - prefect_docker.deployments.steps.build_docker_image:
          id: build-image
          requires: "prefect-docker>=0.4.0"
          image_name: "<http://ghcr.io/org/flow-a|ghcr.io/org/flow-a>"
          tag: "{{ $GITHUB_SHA }}"
    work_pool:
      name: my-k8s
      job_variables:
        image: "{{ build-image.image }}"
        namespace: "prefect"

  - name: flow-b
    entrypoint: flows/flow_b.py:flow
    build:
      - prefect_docker.deployments.steps.build_docker_image:
          id: build-image
          requires: "prefect-docker>=0.4.0"
          image_name: "<http://ghcr.io/org/flow-b|ghcr.io/org/flow-b>"
          tag: "{{ $GITHUB_SHA }}"
    work_pool:
      name: my-k8s
      job_variables:
        image: "{{ build-image.image }}"
        namespace: "prefect"

In CI, make sure the Docker integration is available when deploying:

uv run --with 'prefect[docker]' prefect deploy --all

Common gotchas - One build step id for many deployments: If you have a single top-level build step like

id: build-image

and you reference

{{ build-image.image }}

in multiple deployments, they’ll all get the same image. Use per-deployment build steps or per-deployment env vars instead. - One env var for many deployments: A single

IMAGE

env var will be the same for all deployments in that process. Use

FLOW_A_IMAGE

,

FLOW_B_IMAGE

, etc., or run

prefect deploy --name <deployment>

multiple times with different env vars for each. If you share a snippet of your current

prefect.yaml

and how you invoke

prefect deploy

in GitHub Actions, I can pinpoint exactly why it’s not working and tailor the fix. Helpful docs: - Prefect YAML reference: Define deployments with prefect.yaml - Customize job variables: Customize job variables - Kubernetes worker guide: Run flows on Kubernetes