Kling 4.0 vs WAN 3.0: Specs, References, Edit and Extend Compared

WAN 3.0 from Alibaba’s Tongyi Lab and Kling from Kuaishou are both available on WaveSpeedAI, and with Kling 4.0 live since October 2026 the two now overlap on several headline features: clips up to 30 seconds, video edit and video extend.

This is a spec-level comparison, not a benchmark. We have not run head-to-head quality tests, so the guide below compares what each model accepts and how it is billed, then shows how to test them against each other on your own material. Kling 4.0 is available through the Kling 4.0 API page, the Kling 4.0 collection and the Video Generator.

Quick answer

Pick WAN 3.0 when you want 480p drafts alongside 720p and 1080p, 4:3 or 3:4 aspect ratios, or up to 5 audio references.
Pick Kling 4.0 when you want 21:9, up to 10 images, 5 videos and 3 audio clips as references, keyframes between first and last frame, or an edit that accepts source video up to 30 seconds.
Pick Kling 3.0 or O3 when you need native 4K delivery, up to six shots in one request, or reusable characters through Kling Elements.
Test both on your own prompts; the specs below tell you what each accepts, not which result you will prefer.

Specs side by side

WAN 3.0 and Kling 3.0 values reflect live model parameters on WaveSpeedAI. Kling 4.0 values come from its WaveSpeedAI model pages.

	WAN 3.0 / WAN 3.0 Prime	Kling 4.0	Kling 3.0 / O3
Clip length	2 to 30 seconds	3 to 30 seconds (whole seconds, default 5)	3 to 15 seconds
Resolution	480p, 720p, 1080p	720p (default), 1080p	Standard, Pro, and native 4K tiers
Aspect ratios	16:9, 9:16, 1:1, 4:3, 3:4	16:9, 9:16, 1:1, 21:9	16:9, 9:16, 1:1
Native audio	Yes, on by default	Yes, synchronized stereo audio	Yes (`sound`) on Standard, Pro, 4K
First and last frame	Yes (`last_image` on image-to-video)	Yes (`image`, `last_image`, plus optional `keyframes`; at most 10 images in total)	Yes (`end_image` on Standard, Pro, and 4K image-to-video)
References	Up to 10 images, 5 videos, and 5 audio clips	Up to 10 images, 5 videos, and 3 audio clips (15 files at most; audio 5 to 30 s each)	Up to 3 Kling Elements (3.0 image-to-video and O3); O3 reference-to-video adds up to 7 images and a video
Multi-shot in one request	Single continuous generation	Single request per clip	Up to 6 shots via `multi_prompt`
Edit existing video	Yes, up to 15 seconds of source	Yes, 3 to 30 seconds of source (longer is trimmed to 30)	Yes (O3)
Extend existing video	Yes, 2 to 30 seconds added	Yes, 3 to 30 seconds added	No dedicated endpoint
Faster tier	WAN 3.0 Prime	None on WaveSpeedAI	Kling 3.0 Turbo

Where the two differ

Resolution and aspect ratio

WAN 3.0 spans 480p, 720p and 1080p. Kling 4.0 on WaveSpeedAI is 720p (default) or 1080p, with no 4K tier. For aspect ratio, WAN 3.0 adds 4:3 and 3:4, while Kling 4.0 text-to-video adds 21:9. On Kling 4.0 image-to-video you can omit the aspect ratio to follow the first frame, and reference-to-video defaults to auto, following the references.

References

WAN 3.0 reference-to-video accepts up to 10 reference images, 5 reference videos and 5 reference audio clips in one request. Kling 4.0 reference-to-video accepts up to 10 images, 5 videos and 3 audio clips, 15 files at most, with audio between 5 and 30 seconds each. In the Kling 4.0 prompt you point at the files as @Image1, @Video1 and @Audio1 in the order given, and keep_original_sound keeps a reference video’s own sound instead of generating new audio.

First frame, last frame and in-between frames

Both models take a first frame and a last frame on image-to-video. Kling 4.0 image-to-video also accepts keyframes, in-between frames given in order, with image, last_image and keyframes together limited to 10 images.

Edit and extend

Both families have edit and extend endpoints. WAN 3.0 video-edit revises up to 15 seconds of existing footage with image and audio references, and its video-extend endpoint adds 2 to 30 seconds, optionally ending on a target frame.

Kling 4.0 video-edit takes a 3 to 30 second source (longer is trimmed to 30 seconds) and a prompt describing the change. It keeps the scene, subject and motion, keeps the original sound by default, accepts up to 10 reference images, and follows the input video’s length and aspect ratio unless you set duration or aspect_ratio. Kling 4.0 video-extend generates a new 3 to 30 second segment continuing from the last frame and returns the original and the new segment joined; an optional last_image gives it a target last frame to move toward. Only the new segment is generated and charged.

In our test, a 5 second coastal-road clip of a red vintage convertible was edited into a snowy winter evening: the same car, driver and motion carried through, 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, producing a 10 second clip with audio. These are two single runs, not a benchmark.

Multi-shot and characters

Kling 3.0 accepts up to six shots in one request through multi_prompt, with automatic or custom shot splitting. WAN 3.0 and Kling 4.0 generate one clip per request, so a shot list means one request per shot plus assembly. Kling Elements let you register a character once and reuse it across clips on Kling 3.0 image-to-video and Kling O3.

Native 4K

Kling 3.0 and Kling O3 both have 4K tiers with the same inputs as Standard and Pro. If your deliverable is 4K, those are the models that render it natively on WaveSpeedAI; Kling 4.0 and WAN 3.0 stop at 1080p.

Kling’s own announcements describe Kling 4.0 with additional capabilities such as 4K and HDR (see the Kling 4.0 overview and its early-access Flash announcement of 28 September 2026). What this article describes is the Kling 4.0 offering on WaveSpeedAI, which is exactly the set of endpoints and limits listed above.

Pricing

Both families are billed per use on WaveSpeedAI, with no subscription. At the time of writing:

Kling 4.0 text-to-video, image-to-video and video-extend cost $0.157 per second at 720p and $0.235 per second at 1080p. A 5 second 720p clip is $0.785; a 10 second 1080p clip is $2.35.
Kling 4.0 video-edit, and reference-to-video when you pass reference videos, are billed per counted second: the output length plus the longer of the output length and the total reference-video length, each capped at 30 seconds. That is $0.094 per counted second at 720p and $0.141 at 1080p. For example, when a video edit follows a 5 second input, it is $0.94 at 720p and $1.41 at 1080p. Reference audio is not charged, and reference-to-video without reference videos is billed at the plain per-second rates above.
WAN 3.0 rates depend on duration, resolution and audio.

Because the price basis differs (per generated second versus counted seconds that include reference video), compare cost per approved clip rather than per-second rates alone. See the Kling 4.0 API page and the WAN 3.0 API page for current prices.

How to test Kling 4.0 and WAN 3.0 against each other

Build a matched prompt set. Same prompts, same reference assets, same aspect ratio and the same target length wherever both models support it. Stay within the shared options: 16:9, 9:16 or 1:1, and 720p or 1080p.
Cover the jobs where they differ. Long single takes up to 30 seconds, reference-matched scenes with audio, edits of an existing clip, extensions that continue a clip, and any 21:9, 4:3 or 3:4 deliverables.
Keep settings identical. Same duration, same resolution, same source clips for edit and extend.
Score production outcomes. Identity drift, motion quality, audio sync, how often a clip is approved on the first try, and total cost per approved clip.
Route by job type rather than choosing one model for everything.

The WaveSpeedAI Video Generator has Generate, Reference, Edit and Extend modes for Kling 4.0 and runs WAN 3.0 as well, which is the fastest first pass before scripting the full comparison.

One integration for both

WAN 3.0 and Kling share the same WaveSpeedAI request flow: POST https://api.wavespeed.ai/api/v3/<model id>, receive a prediction ID, then poll https://api.wavespeed.ai/api/v3/predictions/<id>/result. A routing table keeps the choice in configuration:

ROUTES = {
    "long_take": "alibaba/wan-3.0/text-to-video",
    "long_take_kling": "kwaivgi/kling-v4.0/text-to-video",
    "reference_match": "alibaba/wan-3.0/reference-to-video",
    "reference_match_kling": "kwaivgi/kling-v4.0/reference-to-video",
    "edit_kling": "kwaivgi/kling-v4.0/video-edit",
    "extend_kling": "kwaivgi/kling-v4.0/video-extend",
    "shot_list": "kwaivgi/kling-v3.0-pro/text-to-video",
    "delivery_4k": "kwaivgi/kling-v3.0-4k/text-to-video",
}

Bottom line

WAN 3.0 and Kling 4.0 now cover the same ground on WaveSpeedAI: 30-second clips, references, edit and extend. WAN 3.0 adds a 480p tier, 4:3 and 3:4, and a fifth audio reference slot; Kling 4.0 adds 21:9, keyframes, and edit sources up to 30 seconds. Kling 3.0 and O3 remain the choice for native 4K, scripted multi-shot scenes and recurring characters. Keep them in your routing table and decide on your own prompts.

FAQ

Is Kling 4.0 better than WAN 3.0?

We have not benchmarked them head to head, so we do not claim either produces better video. Both are live on WaveSpeedAI with clips up to 30 seconds, video edit and video extend. They differ in concrete specs: WAN 3.0 reaches 1080p with 480p and 720p options, Kling 4.0 offers 720p and 1080p; Kling 4.0 reference-to-video takes up to 10 images, 5 videos and 3 audio clips, WAN 3.0 up to 10 images, 5 videos and 5 audio clips; Kling 4.0 adds a 21:9 aspect ratio, WAN 3.0 adds 4:3 and 3:4. Run both on your own prompts.

What is the difference between WAN 3.0 and WAN 3.0 Prime?

WAN 3.0 Prime is the accelerated variant of WAN 3.0. Both expose the same tasks and inputs on WaveSpeedAI: text-to-video, image-to-video, reference-to-video, video editing, and video extension.

Which model should I use for 4K output?

The native 4K tiers on WaveSpeedAI are Kling 3.0 and Kling O3. Kling 4.0 on WaveSpeedAI is 720p or 1080p, and WAN 3.0 goes up to 1080p, so neither has a 4K tier there.

What video lengths do Kling 4.0 and WAN 3.0 support?

Both generate clips up to 30 seconds. Kling 4.0 accepts 3 to 30 seconds in whole seconds (default 5) on every endpoint; WAN 3.0 accepts 2 to 30 seconds.

How is Kling 4.0 priced compared with WAN 3.0?

Both are billed per use on WaveSpeedAI, with no subscription. Kling 4.0 text-to-video, image-to-video and video-extend are priced per second of generated video, at the time of writing $0.157 per second at 720p and $0.235 per second at 1080p. Kling 4.0 video-edit and reference-to-video with reference videos are billed per counted second instead. Check the Kling 4.0 API page and the WAN 3.0 API page for current rates.

Can I call WAN 3.0 and Kling through the same API?

Yes. On WaveSpeedAI both use the same submit-and-poll request pattern, so switching between them is a model path change plus a few model-specific parameters.