Kling

Kling is Kuaishou's video generation model family. Like other video models, you create a task via POST /v1/videos and then poll for the result (OpenAI Video style responses). Model names map 1:1 to the upstream (kling-v3, kling-v2-6, etc.) — the platform provides no aliases.

Endpoints

• Create task: POST $URL/v1/videos
• Query task: GET $URL/v1/videos/{task_id}

Supported Models

Model Name	Notes	Resolutions
kling-v3	3.0 flagship	720p / 1080p / 4k
kling-v2-6	2.6 standard	720p / 1080p
kling-v2-5-turbo	2.5 fast	720p / 1080p
kling-v2-1-master	2.1 master	720p / 1080p
kling-v2-master	2.0 master	720p / 1080p

• The resolution tier is controlled by mode: std → 720P, pro → 1080P, 4k → 4K; 4K is currently only available on kling-v3 — other models return 400 for 4k;
• kling-v3 and kling-v2-6 support audio generation (metadata.sound); the audio tier is billed higher than the silent tier;
• The 1.x series (kling-v1 / kling-v1-6) is not available.

Parameters

Parameter	Type	Required	Description
model	string	Yes	Model name from the table above
prompt	string	Yes	Video content description
resolution 🆕	string	No	Resolution tier (recommended): "720p"(default) / "1080p" / "4k" (4k only for kling-v3). Conflicting mode returns 400
mode	string	No	Resolution tier (legacy, compatible): std=720p / pro=1080p / 4k, default std
audio 🆕	string	No	Audio generation (official style): "off"(default) / "native". Supported on kling-v3 / kling-v2-6 with higher audio-tier pricing
metadata.sound	string	No	Audio generation (legacy, compatible): "on" / "off", equivalent to audio
aspect_ratio 🆕	string	No	Aspect ratio (official style): 16:9 / 9:16 / 1:1, default 16:9
settings 🆕	object	No	Official settings container fully compatible: settings.resolution / settings.audio / settings.duration / settings.aspect_ratio
seconds	string	No	Output duration in seconds, e.g. "5", "10". Default "5"
size	string	No	Aspect ratio: 16:9 / 9:16 / 1:1 (dimensions like 1920x1080 are also accepted and mapped to a ratio). Default 16:9
image	string	No	Reference image URL (image-to-video)
images	array	No	Reference image URLs
metadata.sound	string	No	Generate audio: "on" / "off" (kling-v3 / kling-v2-6 bill the audio tier higher; only these two models support audio)

Request Example

Text-to-video

curl

curl $URL/v1/videos \
  -H "Authorization: Bearer $YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kling-v3",
    "prompt": "A cat wearing sunglasses walking on a neon street",
    "mode": "std",
    "seconds": "5",
    "metadata": { "sound": "on" }
  }'

Python

import time
import requests

url = "$URL/v1/videos"
headers = {"Authorization": "Bearer YOUR_API_KEY", "Content-Type": "application/json"}
payload = {"model": "kling-v3", "prompt": "A cat on a neon street", "mode": "std", "seconds": "5", "metadata": {"sound": "on"}}

task = requests.post(url, json=payload, headers=headers).json()
task_id = task["id"]

while True:
    result = requests.get(f"{url}/{task_id}", headers=headers).json()
    if result["status"] in ("completed", "failed"):
        break
    time.sleep(8)

print(result)

Image-to-video

curl $URL/v1/videos \
  -H "Authorization: Bearer $YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kling-v3",
    "prompt": "Make the person in the frame turn and walk toward the camera",
    "image": "https://example.com/ref.png",
    "mode": "std",
    "seconds": "5"
  }'

Response Fields

Both create and query return the same structure (OpenAI Video style):

Field	Type	Description
id	string	Task ID
model	string	Model actually used
status	string	Task status (see below)
created_at	integer	Creation timestamp (seconds)
seconds 🆕	string	Requested generation duration in seconds (consistent with billing)
video_url 🆕	string	Video URL after completion (only when completed, same as metadata.url)
resolution 🆕	string	Billed resolution tier: 720p / 1080p / 4k (echoed from submission, consistent with billing)
mode 🆕	string	Tier echo: std / pro / 4k
aspect_ratio 🆕	string	Aspect ratio echo: 16:9 / 9:16 / 1:1
audio 🆕	bool	Whether audio was generated (consistent with billing audio tier)
error.message	string	Failure reason (only when failed)

Task Status

Status	Meaning	Action
queued	Queued	Keep polling
in_progress	Generating	Keep polling
completed	Success	Get the video URL from the response and download it soon
failed	Failed	Check error.message

Poll every 5-10 seconds

Video URLs expire after a while — download and save them as soon as the task completes.

Response Example

{ "id": "task_xxxxxxxx", "model": "kling-v3", "status": "in_progress", "created_at": 1785000000 }

After completion:

{ "id": "task_xxxxxxxx", "model": "kling-v3", "status": "completed", "created_at": 1785000000, "video_url": "https://.../kling-v3-xxxx.mp4" }

Native Protocol Passthrough (advanced)

Raw Kling-format passthrough (the platform handles auth / channel selection / task management, fields pass through verbatim):

Operation	Method & Path
Text-to-video	POST $URL/kling/v1/videos/text2video
Image-to-video	POST $URL/kling/v1/videos/image2video
Query task	GET $URL/kling/v1/videos/{action}/{task_id}

Submissions still go through platform task management (polling) and return a platform task_id. The body is the native Kling format (model_name / duration / mode pass through directly). The omni reference path (/kling/v1/videos/omni-video) requires omni-series models, which are not available yet.

curl $URL/kling/v1/videos/text2video \
  -H "Authorization: Bearer $YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model_name": "kling-v3",
    "prompt": "A cat wearing sunglasses walking on a neon street",
    "mode": "std",
    "duration": "5"
  }'

FAQ

Q: How long does a task take?

Usually 1-3 minutes, longer for complex tasks. Poll the query endpoint; tasks stuck over 20 minutes are auto-failed.

Q: Do video links expire?

Yes. Download and store the video soon after generation completes.