JANCTION Render GitHub Action

Render a Blender scene, a bpy script or a 3D file (GLB, glTF, FBX, USD, OBJ) on JANCTION Render's cloud GPUs from a workflow: a preview image on every push, a turntable or an animation MP4 on a release. No Blender or GPU on the runner.

The action runs the janction-render command line tool (PyPI), the same API the MCP server uses. It estimates the job first and stops before anything starts when the job could cost more than max-cost-yen; the same cap is sent with the job, so the service refuses it before any credit is held.

Use

1. Get an API key: curl -s -X POST https://render.janction.jp/v1/keys returns a new api_key that starts with free GPU time (no sign-up). A key the CLI already uses is in ~/.janction-render.json.
2. Add it as a repository secret named JANCTION_RENDER_API_KEY.
3. Add a workflow:

name: render
on:
  push:
    paths: ["scenes/**"]
jobs:
  preview:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: JasmyLab-JANCTION/render-action@v1
        id: render
        with:
          api-key: ${{ secrets.JANCTION_RENDER_API_KEY }}
          scene-file: scenes/product.blend
          kind: preview
          frames: "1-24"            # up to 4 frames tiled in one image
          max-cost-yen: 0           # only what the free GPU time covers
      - uses: actions/upload-artifact@v4
        with:
          name: render-preview
          path: render-output

A final MP4 on release:

      - uses: JasmyLab-JANCTION/render-action@v1
        with:
          api-key: ${{ secrets.JANCTION_RENDER_API_KEY }}
          scene-file: scenes/turntable.blend
          kind: final
          frames: "1-96"
          width: 1920
          height: 1080
          output: mp4
          max-cost-yen: 200
          timeout-seconds: 1800

Inputs

input	default
api-key	(required)	Repository secret with the API key.
scene-file	(required)	Path to a .blend, a .py bpy script or a 3D file, or an https link. Files a .blend references by relative path are sent with it.
kind	preview	preview (up to 4 frames in one image, up to 1280x720) or final.
frames	1	Preview: 1, 1-24 or 1,8,16. Final: a range such as 1-48 (at most 240 per job).
width, height	1280, 720	Final up to 1920x1080.
samples	16 / 128	Cycles samples per pixel (preview / final).
output	auto	Final: mp4, png, exr, webm, gif or webp.
max-cost-yen	100	The job is refused before it starts if it could cost more (JPY). 0 = free GPU time only.
timeout-seconds	900	Cancels the job after this long (only frames already rendered are charged).
output-dir	render-output	Where the files are written.
allow-fork-pull-requests	false	pull_request_target runs from forks are refused unless this is true.
cli-version	pinned	Version of janction-render on PyPI (each release of this action pins a tested one).

Outputs: job-id, files (JSON array), gpu-seconds, charged-yen. The job summary shows the same.

Safety

• The key is masked in the log and passed to the tool through the environment, never on a command line.
• Without a key the action fails; it never creates a key on its own.
• Pull requests from forks do not get secrets on pull_request; on pull_request_target the action refuses forks unless allow-fork-pull-requests: true.
• A rerun of the same workflow attempt reuses the job (idempotency key from the run id, the attempt and the scene), so a retried step never renders or charges twice. A new attempt ("Re-run jobs") renders again.
• The cap is checked twice: by the estimate before the job is submitted, and by the service when it receives the job.

Pricing and free GPU time: https://render.janction.jp/pricing. Questions: https://render.janction.jp/support.

Source and versions

This repository is published from the integrations/github-action folder of JasmyLab-JANCTION/janction-render, where the command line tool and the MCP server live. v1 follows the latest 1.x release; pin a full tag (for example v1.0.0) to stay on one. Issues: https://github.com/JasmyLab-JANCTION/janction-render/issues