Kling AI Batch Processing
Overview
Generate multiple videos efficiently using controlled parallelism, rate-limit-aware submission, progress tracking, and result collection. All requests go through https://api.klingai.com/v1.
Batch Submission with Rate Limiting
import jwt, time, os, requests
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 submit_batch(prompts, model="kling-v2-master", duration="5",
mode="standard", max_concurrent=3, delay=2.0):
"""Submit batch with controlled concurrency and pacing."""
tasks = []
active = []
for i, prompt in enumerate(prompts):
# Wait if at concurrency limit
while len(active) >= max_concurrent:
active = [t for t in active if not check_complete(t["task_id"])]
if len(active) >= max_concurrent:
time.sleep(5)
response = requests.post(f"{BASE}/videos/text2video", headers=get_headers(), json={
"model_name": model,
"prompt": prompt,
"duration": duration,
"mode": mode,
})
data = response.json()["data"]
task = {"task_id": data["task_id"], "prompt": prompt, "index": i}
tasks.append(task)
active.append(task)
print(f"[{i+1}/{len(prompts)}] Submitted: {data['task_id']}")
time.sleep(delay) # pace requests
return tasks
def check_complete(task_id):
r = requests.get(f"{BASE}/videos/text2video/{task_id}", headers=get_headers()).json()
return r["data"]["task_status"] in ("succeed", "failed")
Collect Results
def collect_results(tasks, timeout=600):
"""Wait for all tasks and collect results."""
results = {}
start = time.monotonic()
while len(results) < len(tasks) and time.monotonic() - start < timeout:
for task in tasks:
if task["task_id"] in results:
continue
r = requests.get(
f"{BASE}/videos/text2video/{task['task_id']}", headers=get_headers()
).json()
status = r["data"]["task_status"]
if status == "succeed":
results[task["task_id"]] = {
"status": "succeed",
"url": r["data"]["task_result"]["videos"][0]["url"],
"prompt": task["prompt"],
}
elif status == "failed":
results[task["task_id"]] = {
"status": "failed",
"error": r["data"].get("task_status_msg", "Unknown"),
"prompt": task["prompt"],
}
if len(results) < len(tasks):
time.sleep(15)
return results
Async Batch with asyncio
import asyncio
import aiohttp
async def async_batch(prompts, max_concurrent=3):
"""Async batch processing with semaphore-controlled concurrency."""
semaphore = asyncio.Semaphore(max_concurrent)
results = {}
async def generate_one(prompt, index):
async with semaphore:
async with aiohttp.ClientSession() as session:
# Submit
async with session.post(
f"{BASE}/videos/text2video",
headers=get_headers(),
json={"model_name": "kling-v2-master", "prompt": prompt,
"duration": "5", "mode": "standard"},
) as resp:
data = (await resp.json())["data"]
task_id = data["task_id"]
# Poll
while True:
await asyncio.sleep(10)
async with session.get(
f"{BASE}/videos/text2video/{task_id}",
headers=get_headers(),
) as resp:
data = (await resp.json())["data"]
if data["task_status"] == "succeed":
results[index] = data["task_result"]["videos"][0]["url"]
return
elif data["task_status"] == "failed":
results[index] = f"FAILED: {data.get('task_status_msg')}"
return
await asyncio.gather(*[generate_one(p, i) for i, p in enumerate(prompts)])
return results
Batch with Callbacks (No Polling)
def submit_batch_with_callbacks(prompts, callback_url):
"""Submit batch with webhook callbacks -- no polling needed."""
tasks = []
for prompt in prompts:
r = requests.post(f"{BASE}/videos/text2video", headers=get_headers(), json={
"model_name": "kling-v2-master",
"prompt": prompt,
"duration": "5",
"mode": "standard",
"callback_url": callback_url,
}).json()
tasks.append(r["data"]["task_id"])
time.sleep(2) # rate limit pacing
return tasks
Cost Estimation Before Batch
def estimate_batch_cost(count, duration=5, mode="standard", audio=False):
credits_map = {(5, "standard"): 10, (5, "professional"): 35,
(10, "standard"): 20, (10, "professional"): 70}
per_video = credits_map.get((duration, mode), 10)
if audio:
per_video *= 5
total = count * per_video
print(f"Batch: {count} videos x {per_video} credits = {total} credits")
print(f"Estimated cost: ${total * 0.14:.2f}")
return total
# Check before submitting
needed = estimate_batch_cost(50, duration=5, mode="standard")
Prerequisites
- An approved batch manifest with a unique batch ID, a bounded count, model, duration, mode, destination, and credit ceiling.
- Prompts and reference media must be synthetic or rights-cleared, and the request must pass the provider's content policy review. Do not submit real people's likenesses, private data, or copyrighted material without documented permission.
- Use a sandbox project and draft/watermarked outputs for the first canary. Store
KLING_ACCESS_KEYandKLING_SECRET_KEYin the approved secret manager; never place them in prompts, source control, or logs.
Instructions
- Validate the manifest before any request: reject missing rights/consent, disallowed content, unapproved destinations, duplicate batch IDs, and a projected credit total above the approved ceiling.
- Run one synthetic canary with the lowest-cost permitted mode. Confirm the model, duration, aspect ratio, callback destination, watermark/draft status, and
contacts_exported=0-style no-export invariant before expanding the batch. - Submit only the approved count with bounded concurrency and pacing. Record opaque task IDs and an idempotency key; never log prompts, source media, callback secrets, or result URLs.
- Poll or receive callbacks with a timeout and a retry budget. Hold successful outputs in quarantine until an owner reviews content policy, rights, quality, and cost results.
- Promote approved outputs to the allowlisted destination, then expire temporary artifacts and access. Keep only a redacted receipt and the rollback reference.
Output
Return a batch receipt containing the opaque batch ID, model/mode/duration, requested and completed counts, success/failure counts, credit estimate and actual, canary result, policy/rights review state, destination class, retention deadline, and rollback/removal action. The receipt must exclude prompts, media, personal data, credentials, and signed URLs.
Error Handling
- Retry only bounded transient transport or rate-limit failures with exponential backoff; do not retry policy refusals, rights failures, authentication failures, or invalid parameters.
- If the credit ceiling, concurrency limit, policy probe, or destination allowlist check fails, stop new submissions and mark the batch paused. Reconcile unknown task states before deciding whether to retry.
- On a failed canary or review, cancel pending tasks where supported, remove quarantined outputs, revoke temporary callback access, and record the redacted removal receipt. Restore the last approved batch configuration rather than silently changing scope.
Examples
For a safe dry run, use batch_id=synthetic-launch-01, 3 synthetic prompts, model=kling-v2-5-turbo, duration=5, mode=standard, destination=sandbox-review, watermark=draft, credits_max=30, and contacts_exported=0. Promote only after the owner records policy=pass, rights=pass, and approval=granted; otherwise remove the canary outputs.