Kling AI CI Integration

Overview

Automate video generation in CI/CD pipelines. Common use cases: generate product demos on release, create marketing videos from prompts in a YAML file, regression-test video quality across model versions.

GitHub Actions Workflow

# .github/workflows/generate-videos.yml
name: Generate Videos
on:
  workflow_dispatch:
    inputs:
      prompt:
        description: "Video prompt"
        required: true
      model:
        description: "Model version"
        default: "kling-v2-master"

jobs:
  generate:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4

      - uses: actions/setup-python@v5
        with:
          python-version: "3.11"

      - name: Install dependencies
        run: pip install PyJWT requests

      - name: Generate video
        env:
          KLING_ACCESS_KEY: ${{ secrets.KLING_ACCESS_KEY }}
          KLING_SECRET_KEY: ${{ secrets.KLING_SECRET_KEY }}
        run: |
          python3 scripts/generate-video.py \
            --prompt "${{ inputs.prompt }}" \
            --model "${{ inputs.model }}" \
            --output output/

      - name: Upload artifact
        uses: actions/upload-artifact@v4
        with:
          name: generated-video
          path: output/*.mp4
          retention-days: 7

CI Generation Script

#!/usr/bin/env python3
"""scripts/generate-video.py -- CI-friendly video generation."""
import argparse
import jwt
import time
import os
import requests
import sys

BASE = "https://api.klingai.com/v1"

def get_headers():
    ak, sk = os.environ["KLING_ACCESS_KEY"], os.environ["KLING_SECRET_KEY"]
    token = jwt.encode(
        {"iss": ak, "exp": int(time.time()) + 1800, "nbf": int(time.time()) - 5},
        sk, algorithm="HS256", headers={"alg": "HS256", "typ": "JWT"}
    )
    return {"Authorization": f"Bearer {token}", "Content-Type": "application/json"}

def main():
    parser = argparse.ArgumentParser()
    parser.add_argument("--prompt", required=True)
    parser.add_argument("--model", default="kling-v2-master")
    parser.add_argument("--duration", default="5")
    parser.add_argument("--mode", default="standard")
    parser.add_argument("--output", default="output/")
    parser.add_argument("--timeout", type=int, default=600)
    args = parser.parse_args()

    os.makedirs(args.output, exist_ok=True)

    # Submit
    r = requests.post(f"{BASE}/videos/text2video", headers=get_headers(), json={
        "model_name": args.model,
        "prompt": args.prompt,
        "duration": args.duration,
        "mode": args.mode,
    })
    r.raise_for_status()
    task_id = r.json()["data"]["task_id"]
    print(f"Task submitted: {task_id}")

    # Poll
    start = time.monotonic()
    while time.monotonic() - start < args.timeout:
        time.sleep(15)
        result = requests.get(
            f"{BASE}/videos/text2video/{task_id}", headers=get_headers()
        ).json()
        status = result["data"]["task_status"]
        elapsed = int(time.monotonic() - start)
        print(f"[{elapsed}s] Status: {status}")

        if status == "succeed":
            video_url = result["data"]["task_result"]["videos"][0]["url"]
            filepath = os.path.join(args.output, f"{task_id}.mp4")
            with open(filepath, "wb") as f:
                f.write(requests.get(video_url).content)
            print(f"Saved: {filepath}")
            return

        if status == "failed":
            print(f"FAILED: {result['data'].get('task_status_msg')}", file=sys.stderr)
            sys.exit(1)

    print("TIMEOUT: generation did not complete", file=sys.stderr)
    sys.exit(1)

if __name__ == "__main__":
    main()

Batch from YAML Config

# video-prompts.yml
videos:
  - name: product-hero
    prompt: "Sleek laptop floating in space with particle effects"
    model: kling-v2-6
    mode: professional
  - name: feature-demo
    prompt: "Dashboard interface morphing between screens"
    model: kling-v2-5-turbo
    mode: standard

import yaml

with open("video-prompts.yml") as f:
    config = yaml.safe_load(f)

for video in config["videos"]:
    task_id = submit_async(video["prompt"], model=video["model"])
    print(f"{video['name']}: {task_id}")

GitLab CI

# .gitlab-ci.yml
generate-video:
  image: python:3.11-slim
  stage: build
  script:
    - pip install PyJWT requests
    - python3 scripts/generate-video.py --prompt "$VIDEO_PROMPT" --output output/
  artifacts:
    paths:
      - output/*.mp4
    expire_in: 7 days
  variables:
    KLING_ACCESS_KEY: $KLING_ACCESS_KEY
    KLING_SECRET_KEY: $KLING_SECRET_KEY

Secret Management

Never put API keys in the workflow YAML or commit them to the repo.

Prerequisites

• A CI environment with Python 3.11+, pinned dependencies, a secret-manager-backed Kling credential, and an explicit per-run credit and concurrency budget.
• A repository-controlled model, duration, destination, and content-policy allowlist. CI fixtures must be synthetic or rights-cleared; never use customer media or real-person likenesses in unattended jobs.
• A private artifact bucket, short retention period, and an approval gate. Automated jobs produce draft, watermarked media only until a named owner approves promotion.

Instructions

1. Validate the workflow and prompt manifest before any API call: require a synthetic/rights-cleared fixture identifier, approved model and duration, permitted destination, and a nonzero but bounded budget.
2. Load credentials only from masked CI secrets, run a policy and consent check, and use a stable manifest hash to prevent duplicate submissions on retries.
3. Submit a single sandbox canary first. Assert that output remains private and watermarked, that no source or prompt is echoed into logs, and that the run has not exceeded its credit or concurrency budget.
4. Require an owner approval artifact before promotion. Publish by immutable output digest to the allowlisted bucket; do not publish directly from a provider URL.
5. On cancellation, policy rejection, budget breach, or failed verification, fail the job closed, revoke temporary access, delete staged artifacts, and restore the prior release manifest. Retain only a redacted receipt.

Output

The job should emit a machine-readable receipt with the run and manifest digests, model, environment, canary status, policy and rights checks, budget usage, approval status, artifact digest, retention deadline, and rollback reference. CI logs may contain task status and timings, but must exclude credentials, source URLs, prompts, face or contact data, and raw provider responses.

Error Handling

Treat authentication failures, policy rejections, unavailable source fixtures, quota or budget errors, and provider timeouts as non-publish failures. Retry only bounded, idempotent polling or transient transport errors; never retry a rejected prompt or blindly resubmit a billable generation. Mark the run for owner review when the provider returns an unknown status, quarantine all artifacts, and use the previous approved manifest for rollback.

Examples

A safe dispatch manifest can be represented as:

fixture: synthetic-product-v4
rights: cleared-for-internal-test
model: kling-v2-6
duration: 5
environment: staging
canary: watermarked-private
budget_credits: 10
publish: false
approval: required

The promotion job should require approval: recorded and an immutable artifact digest; a pull request or scheduled run must never turn an unreviewed live photograph into a public video.

Resources

• API Reference
• GitHub Actions Docs