Kling 4.0 Is Live on WaveSpeedAI: Models, Pricing, and How to Use Them

Kling 4.0 is live on WaveSpeedAI. Five endpoints are available through the API and the browser: text-to-video, image-to-video, reference-to-video, video edit and video extend. Every endpoint makes clips of 3 to 30 seconds at 720p or 1080p with synchronized stereo audio, billed per second of generated video with no subscription. Start from the Kling 4.0 API page, browse the Kling 4 collection, or open the Video Generator.

The five Kling 4.0 endpoints

Model	Model ID	What it takes	Price basis
Text to video	kwaivgi/kling-v4.0/text-to-video	prompt; aspect_ratio 16:9 (default), 9:16, 1:1 or 21:9	Per second of output
Image to video	kwaivgi/kling-v4.0/image-to-video	image as first frame (required); optional last_image and keyframes, up to 10 images in total; optional prompt	Per second of output
Reference to video	kwaivgi/kling-v4.0/reference-to-video	Up to 10 reference images, 5 reference videos and 3 reference audios (15 files at most)	Per second of output, or counted seconds when reference videos are used
Video edit	kwaivgi/kling-v4.0/video-edit	video (3 to 30 s) and a prompt describing the change; optional images	Counted seconds
Video extend	kwaivgi/kling-v4.0/video-extend	video and a prompt; optional last_image as a target final frame	Per second of the new segment

Details worth knowing:

• Duration and resolution. Duration is a whole number of seconds from 3 to 30 (default 5). Resolution is 720p (default) or 1080p.
• Image to video. aspect_ratio is optional; leave it out and the clip follows the first frame. keyframes are in-between frames, given in order.
• Reference to video. Refer to your files in the prompt as @Image1, @Video1, @Audio1, in the order you supplied them. Reference audios must be 5 to 30 seconds each. Set keep_original_sound to keep the reference video’s own sound instead of generating new audio. aspect_ratio defaults to auto and follows the references.
• Video edit. The scene, subject and motion are kept and your prompt describes the change; refer to the clip as @Video1. Original sound is kept by default. Videos longer than 30 seconds are trimmed to 30. The output follows the input’s length and aspect ratio unless you set duration or aspect_ratio.
• Video extend. The model generates a new segment (3 to 30 s, default 5) that continues from the last frame, and returns the original and the new segment joined. Only the new segment is generated and charged.

All five use the same pattern as the other Kling models: POST to https://api.wavespeed.ai/api/v3/<model id>, receive a prediction ID, then poll https://api.wavespeed.ai/api/v3/predictions/<id>/result.

Kling 4.0 pricing on WaveSpeedAI

Prices at the time of writing, in USD. The Kling 4.0 API page always has the current numbers.

Endpoint	720p	1080p
Text to video, image to video, video extend	$0.157 per second	$0.235 per second
Reference to video (no reference video)	$0.157 per second	$0.235 per second
Reference to video (with reference videos), video edit	$0.094 per counted second	$0.141 per counted second

Examples: a 5-second 720p text-to-video clip costs $0.785, and a 10-second 1080p clip costs $2.35.

For reference-to-video with reference videos, and for video-edit, a counted second is the output length plus the longer of the output length and the total reference-video length, each capped at 30 seconds. If the output is as long as the reference videos, that works out to $0.188 per second at 720p and $0.282 per second at 1080p. Reference audios are not charged. For a video edit whose output follows a 5-second input, the cost is $0.94 at 720p and $1.41 at 1080p.

Kling 4.0 vs Kling 3.0, Turbo and O3 in practice

• Length. Kling 4.0 on WaveSpeedAI goes up to 30 seconds per clip. Kling’s VIDEO 3.0 user guide documents 3 to 15 seconds for Kling 3.0.
• Resolution. Kling 4.0 on WaveSpeedAI is 720p or 1080p. The native 4K tiers are on Kling 3.0 4K and Kling O3 4K, so choose those when 4K delivery matters.
• Inputs. Kling 4.0 reference-to-video takes images, videos and audio together, and video edit and video extend work on existing footage. Kling O3 covers similar ground with up to seven reference images, a reference video and Kling Elements.
• Workflow. Prototype on Kling 3.0 Turbo, which is the lowest-cost place to iterate, then render on Kling 4.0 or on the 3.0 and O3 tiers. Because every Kling endpoint uses one request shape, keeping the model ID in configuration makes switching a one-line change.

The rest of the lineup is still available. Current rates are on the Kling 3.0 API page and the Kling O3 API page.

The rest of the Kling lineup

Kling 3.0: the current flagship

The Kling 3.0 family covers text-to-video, image-to-video, and motion control at three quality tiers.

Model	Task	Model path
Kling 3.0 Standard	text-to-video, image-to-video	kwaivgi/kling-v3.0-std/text-to-video, kwaivgi/kling-v3.0-std/image-to-video
Kling 3.0 Pro	text-to-video, image-to-video	kwaivgi/kling-v3.0-pro/text-to-video, kwaivgi/kling-v3.0-pro/image-to-video
Kling 3.0 4K	text-to-video, image-to-video	kwaivgi/kling-v3.0-4k/text-to-video, kwaivgi/kling-v3.0-4k/image-to-video
Kling 3.0 Standard Motion Control	video-to-video	kwaivgi/kling-v3.0-std/motion-control
Kling 3.0 Pro Motion Control	video-to-video	kwaivgi/kling-v3.0-pro/motion-control

Kling 3.0 introduced native 4K output, multi-shot prompts, and improved physical realism. Motion control transfers the movement from a reference clip onto a still image, which is the fastest way to get a specific dance, gesture, or action without prompting for it.

Kling 3.0 Turbo: for iteration

The Turbo variants trade a little fidelity for speed and a lower price, and they support the same single-prompt and multi-shot inputs as the full 3.0 models.

Model	Model path
Kling V3 Turbo Standard	kwaivgi/kling-v3-turbo-std/text-to-video, kwaivgi/kling-v3-turbo-std/image-to-video
Kling V3 Turbo Pro	kwaivgi/kling-v3-turbo-pro/text-to-video, kwaivgi/kling-v3-turbo-pro/image-to-video

Turbo is the right place to develop prompts. Once a prompt and reference set work on Turbo, they transfer to the Standard, Pro, or 4K endpoints unchanged.

Kling O3: reference-to-video and video editing

Kling O3, also marketed as Kling 3.0 Omni, is the unified multimodal branch of the family. Beyond text and image input, it accepts up to seven reference images, a reference video, and reusable Kling Elements for consistent characters across clips. It also edits existing footage.

Tier	Tasks	Model path prefix
O3 Standard	text, image, and reference to video; video edit	kwaivgi/kling-video-o3-std/
O3 Pro	text, image, and reference to video; video edit	kwaivgi/kling-video-o3-pro/
O3 4K	text, image, and reference to video; video edit; video reference	kwaivgi/kling-video-o3-4k/

The 4K tier also has dedicated video-edit and video-reference endpoints, covered in detail in our Kling O3 4K video edit and video reference post.

Kling Elements stay a feature of Kling 3.0 image-to-video and O3; the Kling 4.0 endpoints take reference images, videos and audio directly instead.

Kling 2.6 and earlier

Kling 2.6 Standard and Pro remain available for teams that have tuned prompts against them, along with Kling 2.5 Turbo, Kling LipSync, Kling AI Avatar, and the Kling audio models. See the full list on the Kling 3.0 API page.

What Kling announced vs what you get on WaveSpeedAI

Kling has made its own announcements about Kling 4.0, for example early access to a Flash variant for Ultra Yearly subscribers on September 28, 2026, and a Kling 4.0 overview at kling.ai. Those describe Kling’s own product. What WaveSpeedAI offers is exactly what is listed above: five endpoints, 3 to 30 second clips, 720p and 1080p, with audio. There is no Flash variant, no 4K tier and no Kling Image 4.0 endpoint on WaveSpeedAI. Check the Kling 4.0 API page for the current list.

Try Kling 4.0 without code

The WaveSpeedAI Video Generator runs Kling 4.0 in the browser, next to Seedance 2.5, Google Veo 3.1, WAN 3.0 and other models. Kling 4.0 has four modes:

• Generate for text-to-video and image-to-video.
• Reference for reference-to-video with images, videos and audio.
• Edit to change an existing clip with a prompt.
• Extend to continue a clip from its last frame.

In our test, an edit of a 5-second coastal-road clip with a red vintage convertible, with a prompt to turn it into a snowy winter evening while keeping the car and its motion, returned the same car, driver and motion with snow on the road and cliffs, at 1280x720 with the audio kept. In another test, a 5-second sailboat-at-sunrise clip was extended by 5 seconds: the boat and bay carried on and the camera rose to reveal a harbor, giving a 10-second clip with audio. These are two quick tests, not benchmarks.

Call Kling 4.0 from Python

The sample below submits a text-to-video job and polls for the result. For the other endpoints, change the model ID and the input fields.

import os
import time
import requests

API_KEY = os.environ["WAVESPEED_API_KEY"]
BASE = "https://api.wavespeed.ai/api/v3"

MODEL = "kwaivgi/kling-v4.0/text-to-video"

# Submit once. Do not retry this POST automatically: a retry after an ambiguous
# failure (timeout, 5xx) could create a second paid task.
resp = requests.post(
    f"{BASE}/{MODEL}",
    headers={"Authorization": f"Bearer {API_KEY}", "Content-Type": "application/json"},
    json={
        "prompt": "A tailor pins fabric on a mannequin in a sunlit studio, slow dolly-in, dust in the light beams",
        "duration": 5,
        "resolution": "720p",
        "aspect_ratio": "16:9",
    },
    timeout=60,
)
resp.raise_for_status()
prediction_id = resp.json()["data"]["id"]

# Poll until a terminal status.
TRANSIENT = {429, 500, 502, 503, 504}
failures = 0
while True:
    try:
        poll = requests.get(
            f"{BASE}/predictions/{prediction_id}/result",
            headers={"Authorization": f"Bearer {API_KEY}"},
            timeout=60,
        )
    except requests.RequestException:  # network error: back off and keep polling
        poll = None
    if poll is None or poll.status_code in TRANSIENT:
        failures += 1
        retry_after = poll.headers.get("Retry-After", "") if poll is not None else ""
        time.sleep(int(retry_after) if retry_after.isdigit() else min(2 ** failures, 60))
        continue
    if poll.status_code != 200:
        raise RuntimeError(f"task {prediction_id}: unexpected HTTP {poll.status_code}")
    failures = 0
    result = poll.json()["data"]
    if result["status"] == "completed":
        print(result["outputs"][0])
        break
    if result["status"] in {"failed", "cancelled", "timeout", "deleted"}:
        raise RuntimeError(f"task {prediction_id} {result['status']}: {result.get('error')}")
    time.sleep(2)

Tips:

1. Keep the model ID in configuration, not in code, so moving between Kling 4.0, Kling 3.0 and O3 is a config change.
2. Build a fixed prompt set of 10 to 20 briefs that represent your real workload and run it on Kling 4.0, Kling 3.0 and Seedance 2.5. Compare consistency, motion and cost per approved clip rather than demo reels.
3. Start at 720p (the default, and the lower price) while you work on prompts, then render finals at 1080p.

Related reading

• Kling 3.0 Omni Explained
• Kling O3 4K Adds Video Edit and Video Reference
• Seedance 2.5 vs Kling: Video Generation Compared
• What to Expect from Kling 3.0: A Technical Preview

FAQ

Is Kling 4.0 available on WaveSpeedAI?

Yes. Kling 4.0 has been live on WaveSpeedAI since October 2026 with five endpoints: text-to-video, image-to-video, reference-to-video, video-edit and video-extend. You can call them through the API or use them in the browser Video Generator.

How much does Kling 4.0 cost on WaveSpeedAI?

At the time of writing, text-to-video, image-to-video and video-extend cost $0.157 per second at 720p and $0.235 per second at 1080p, so a 5-second 720p clip is $0.785. Reference-to-video with reference videos and video-edit use a counted-second rule at $0.094 (720p) or $0.141 (1080p) per counted second. Billing is per use with no subscription; check the Kling 4.0 API page for current prices.

What resolutions and durations does Kling 4.0 support on WaveSpeedAI?

All five endpoints produce 3 to 30 second clips (whole seconds, default 5) at 720p (default) or 1080p, with synchronized stereo audio generated alongside the picture. There is no 4K tier on the Kling 4.0 endpoints; native 4K is available on Kling 3.0 4K and Kling O3 4K.

Which Kling model should I use: 4.0, 3.0 Turbo or O3?

Use Kling 4.0 when you want clips up to 30 seconds, reference images, videos and audio in one request, or editing and extending existing footage at 720p or 1080p. Use Kling 3.0 Turbo for low-cost prompt iteration, Kling 3.0 or O3 when you need native 4K, and Kling 3.0 image-to-video or O3 when you rely on Kling Elements. All of them share the same request and polling pattern.

Can I try Kling 4.0 without writing code?

Yes. The WaveSpeedAI Video Generator at wavespeed.ai/video-generator?model=kling-4.0 has four modes for Kling 4.0: Generate, Reference, Edit and Extend.