Updated November 2, 2026. Kling is still rolling out parts of Kling 4.0, so some features below are marked "announced" rather than live. We'll update this page as they ship.
Kling 4.0 is the newest AI video model from Kling AI, the video unit of Chinese tech company Kuaishou. Its headline trick is simple to say and hard to do: one prompt, one pass, up to 30 seconds of video with sound, with no stitching. Kuaishou says it launched on October 15, 2026, after a smaller "Flash" version went to early testers on September 28.
That puts Kling 4.0 in the 30-second club with Alibaba's Wan 3.0 and ByteDance's Seedance 2.5, and well past Google's Veo 3.1, which still makes 8-second clips. It also lands in a market OpenAI has left: the Sora app closed in April and the Sora API was switched off on September 24.
Below: what's live, what's only announced, how it compares, what Kling charges, and how to prompt a 30-second clip that holds together.
Kling 4.0: the short version
- Release date: Kling 4.0 opened to all users on October 15, 2026, per Kuaishou's press release. Kling 4.0 Flash had opened to Ultra yearly subscribers on September 28.
- Length: up to 30 seconds in a single generation, double Kling 3.0's 15-second ceiling.
- Sound: audio moved from one channel to two-channel stereo, with multilingual lip sync, according to launch reports.
- References: "Omni Reference" takes up to 15 references per task (up to 10 images, 5 videos and 7 subjects), plus up to 10 keyframes and prompts of up to 8,000 tokens.
- Not live yet: 4K, 10-bit HDR, video extension to two minutes and the enterprise API were all described as coming, not shipped, in Kuaishou's own release.
- Pricing: Kling sells credit plans from Standard to Ultra. It has not published a Kling 4.0 credit rate on any page we could find.
- On Cleus: Kling 4.0 is available in Imagine for text-to-video, image-to-video, reference and video-edit modes.
How Kling 4.0 rolled out
2026 dates, with the Sora and Veo changes around it. Press play.
OpenAI announces it will discontinue Sora in both the app and the API. At this point Kling 3.0 makes clips of up to 15 seconds.
Sources: Kuaishou press release, OpenAI Help Center, TechNode, Google Gemini API docs, SandBase, Grainzilla.
Kling 4.0 release date: how the rollout went
Kling staged the launch over about three weeks, which is why some guides say "coming soon" and others say "live".
September 28: Kling 4.0 Flash early access
Kling first opened Kling 4.0 Flash, a faster, lighter version, to a limited group of users on September 28. Coverage of Kling's own posts says the group was Ultra yearly subscribers. Testers reported Flash clips topping out at about 20 seconds at 720p in 8-bit SDR (standard dynamic range, the normal colour depth of most web video). Kling never published a full Flash spec sheet, so treat those numbers as tester-reported.
October 15: full launch for all users
Kuaishou's press release, issued in Hong Kong on October 16, says Kling 4.0 "officially launched" on October 15. It was opened to all users on Kling's site. Some outlets date the launch to October 19, which is when a few of them wrote it up. The company's own release is the source to trust here.
October 21 onward: the API
The press release says API access for developers and enterprises "will soon be made available." Reports from October 21 say Kling's official account then announced the API was rolling out. When we loaded Kling's public developer price table on November 2, it still listed only Kling 3.0-series models. So if you need Kling 4.0 through an API, check access on your account before you build on it.
What's new in Kling 4.0 (live vs announced)
Here is every headline feature, sorted by whether you can use it today. "Live" means Kuaishou says it shipped and at least one independent test or report confirms it. "Announced" means Kuaishou says it is coming.
| Feature | Status | What we know |
|---|---|---|
| Up to 30 s in one pass | Live | Kuaishou's headline claim. An independent API test on October 23 produced a 30.04-second clip with no cuts. |
| Two-channel stereo audio | Live | Reported at launch; independent tests found stereo on by default. |
| Multilingual lip sync | Live (claimed) | Kling claims more than ten dialogue languages and "more accurate" lip sync. Not independently benchmarked yet. |
| Omni Reference (15 refs) | Live | Up to 10 images, 5 videos and 7 subjects per task. The API version tested on Oct 23 took 10 images and 5 videos. |
| 10 keyframes | Live in Kling's app | The API tested on Oct 23 exposed only a start frame and optional end frame. |
| Reference-to-video and video edit | Live | Both showed up as working endpoints in independent tests. |
| 1080p and 21:9 | Live in tests | Oddly, the press release lists 1080p and 21:9 under "will also support", but testers got 1080p by default and a 21:9 option. |
| Kling MCP and CLI | Live | Lets AI agents call Kling features like Omni Reference, per the release. |
| 4K output | Announced | Not in the API tested on Oct 23, which offered 720p and 1080p only. |
| 10-bit HDR | Announced | HDR means a wider range of brightness and colour. Test files were 8-bit. |
| Extension to 2 minutes | Announced | Kuaishou says it "will soon support" multiple extensions up to two minutes. |
| Enterprise API | Rolling out | Reported rolling out from Oct 21; not on Kling's public price table as of Nov 2. |
Longer clips without cuts
Most AI video models still make 5 to 10 seconds at a time, so a 30-second scene meant stitching clips together while faces, clothes and lighting drifted between them. One pass keeps the whole scene consistent.
Steadier fast action and camera moves
Kuaishou says Kling 4.0 handles large motion and complex camera work more steadily. It names push-ins, pull-outs, pans, trucking (moving the camera sideways), tracking (following a subject) and orbit shots. These are company claims. One independent tester saw a clean orbit shot, but there are no public benchmarks yet.
Stereo sound and lip sync
Kling 3.0 already made sound. Kling 4.0 moves to two-channel stereo, so a car passing left to right can actually pan. Launch coverage says dialogue and lip sync cover more than ten languages, including English, Japanese, Korean, Spanish and German. If you need lip sync in a less common language, test it yourself before you rely on it.
Omni Reference and keyframes
Omni Reference is Kling's name for feeding the model your own material. Per the release, it accepts subject photos, three-view character sheets (front, side and back of one character), grid images, storyboards and wireframes. It also takes reference videos for a performance, an action, a camera move or pacing. That includes depth-map videos (greyscale clips that encode distance) and "white-model" videos (plain untextured 3D renders), which pin down motion precisely. Keyframes are still images you place at points in the clip so the model knows where the shot has to be at that moment.
Video edit and "creative recreation"
The edit mode takes an existing clip and changes it: Kuaishou mentions swapping products, characters or other assets inside a reference video. Reference video input is capped at 5 clips totalling 30 seconds, per launch coverage.
Kling 4.0 vs Veo 3.1, Wan 3.0 and Seedance 2.5
The three numbers that matter most when you pick a video model are how long one clip can be, how sharp it is, and whether it makes its own sound. Here's how the current models stack up on those, using only figures from the companies' docs and launch coverage.
AI video models by longest single clip
Tap a row for details.
- Up to 30 s in one pass. 1080p live; 4K and 10-bit HDR announced. Two-channel stereo audio with lip sync.
Single-generation length from company docs and launch coverage. Sora 2 is left out: OpenAI discontinued the Sora API on Sep 24, 2026.
A few things the table can't show:
- Veo 3.1 makes the shortest clips but offers 4K output today on its top tier (Seedance 2.0 also lists a 4K option). Google's docs let you extend a Veo clip by 7 seconds at a time, up to 37 seconds total on its enterprise platform. Google also shut down the Veo 3.1 Preview models in the Gemini API on October 22, 2026, pointing developers to its newer Gemini Omni Flash model or the general-release Veo models.
- Wan 3.0 from Alibaba launched on August 24, 2026 with 30-second clips at up to 1080p and sound generated in the same pass. Unlike earlier Wan models, its weights are closed. Dataconomy reported API pricing of $0.20 per second at 1080p.
- Seedance 2.5 from ByteDance runs 4 to 30 seconds at up to 1080p with sound on by default, per published API docs. Its older sibling Seedance 2.0 tops out at 15 seconds but lists a 4K option. ByteDance's own spec page for 2.5 didn't turn up in our research, so treat its numbers as provisional.
- Sora is out of the picture. OpenAI announced the shutdown on March 24, 2026, closed the Sora app and web on April 26, and switched off the Sora API on September 24. It has not named a replacement video model.
There are no independent head-to-head benchmarks for Kling 4.0 yet. On paper, its edge is control: 15 references, 10 keyframes and an edit mode in one model. Veo's edge is 4K. Wan 3.0 and Seedance 2.5 match Kling on length.
Kling 4.0 pricing
Kling sells monthly credit plans on its own site. As of mid-October, a pricing guide listed these (prices shown to new subscribers, then the renewal price):
| Plan | First month | After that | Yearly | Credits / month |
|---|---|---|---|---|
| Standard | $6.99 | $8.80 | $79.20 | 660 |
| Pro | $25.99 | $32.56 | $293.04 | 3,000 |
| Premier | $64.99 | $80.96 | $728.64 | 8,000 |
| Ultra | $127.99 | $159.99 | $1,429.99 | 26,000 |
Prices change with promotions and region, so check Kling's membership page before you pay. Kling 4.0 Flash early access came with the Ultra yearly plan.
What Kling hasn't published is the number you actually need: how many credits a Kling 4.0 second costs. One launch analysis cited third-party estimates of 15 credits per second at 1080p and 10 at 720p, which would make a 30-second 1080p clip 450 credits. That figure is unverified, so don't plan a budget around it.
How to use Kling 4.0: prompting 30-second clips
A 30-second clip is not a 5-second clip stretched out. If you write one line, the model has to invent 25 seconds of story, and that's where drift and random cuts creep in. You'll get more consistent results if you do the directing yourself. Kling 4.0 accepts prompts of up to 8,000 tokens, so there's room to spell out every beat.
Build a 30-second Kling 4.0 prompt
Five steps from idea to a clip that holds together.
Open with one line for the whole clip: film look, lighting, colour, and the overall sound bed. This stops the style drifting between beats.
A worked example
Here's how a 30-second prompt can look when it's broken into beats:
Style: 35mm film look, warm late-afternoon light, shallow depth of field. Stereo street ambience throughout. 0–8 s: Wide establishing shot of a rain-wet night market in Taipei. Slow push-in toward a noodle stall. Steam rises, distant scooters pan left to right. 8–18 s: Medium shot. The cook, a woman in her 50s in a red apron, tosses noodles in a wok, flames flaring. Camera trucks right to follow her hands. 18–26 s: Close-up. She looks up at the camera and says in Mandarin, "Last bowl of the night." Lip-synced, warm tone. 26–30 s: Slow pull-out to a wide shot as the stall lights switch off one by one.
Tips that matter most at 30 seconds
- Write in beats with timestamps. Three to five beats is plenty for 30 seconds. Each beat gets one camera move and one action.
- Use real camera language. "Push-in", "pull-out", "pan", "truck right", "tracking shot", "orbit" and "close-up" are the moves Kuaishou says the model was tuned on.
- Lock the character with references. A three-view character sheet (front, side, back) does more for consistency across 30 seconds than any amount of description.
- Use a reference video for hard motion. If you need a specific dance, fight move or camera path, show it with a reference clip instead of describing it.
- Remember references eat into the 30-second cap. Reference and edit video inputs share a 30-second combined limit, so long reference clips leave less room.
- Put dialogue in quotes and name the language. Short lines, one speaker at a time, work best for lip sync.
- Describe the sound. Stereo is only useful if the prompt says where sounds come from: "traffic passing left to right", "rain on a tin roof overhead".
- Draft short, then go long. Test your look and characters with a quick, low-resolution clip (or Flash, if you have it) before spending on a full 30-second 1080p render.
Using Kling 4.0 on Cleus
Cleus is an uncensored AI platform with chat, AI characters and Imagine for images and video. Kling 4.0 is one of the video models in Imagine:
- Text-to-video: write the prompt, pick a length from 3 to 30 seconds.
- Image-to-video: start from your own picture or one you made in Imagine.
- Reference mode: add reference images or clips to keep characters and style consistent.
- Video edit: change an existing clip instead of starting over.
- Output: 720p or 1080p, with sound, in 16:9, 9:16, 1:1 or 21:9.
You don't need a separate Kling subscription. Sign up, open Imagine and choose Kling 4.0 from the model list. For more of what launched this autumn, see our roundup of new AI models released in October 2026.
Kling 4.0 quiz
Six questions. How closely were you reading?
How long can a single Kling 4.0 generation be?
Kling 4.0 FAQ
When was Kling 4.0 released?
Kuaishou says Kling 4.0 officially launched on October 15, 2026, and announced it in a press release dated October 16. A lighter version, Kling 4.0 Flash, opened earlier, on September 28, to a limited group of Ultra yearly subscribers. Some sites give October 19, but that is when they covered it, not the launch date.
How long can Kling 4.0 videos be?
Up to 30 seconds in one generation, according to Kuaishou, and an independent test produced a 30.04-second clip without cuts. The shortest length is reported as 3 seconds. Kuaishou has also announced extensions up to two minutes, but that feature was still listed as coming soon in its release and isn't live yet.
Is Kling 4.0 free?
Kling's Basic plan is free and comes with some free credits, per pricing guides. But Kling hasn't said how many credits a Kling 4.0 clip costs, or whether free credits cover it, and early Flash access was limited to the Ultra yearly plan. For regular use, plan on a paid Kling plan or a platform that includes Kling 4.0, such as Cleus Imagine.
Does Kling 4.0 support 4K and HDR?
Not yet. Kuaishou's release lists 4K and up to 10-bit HDR, but describes them as coming rather than shipped. An independent API test on October 23 found only 720p and 1080p output and 8-bit files. Treat 4K and HDR as announced features until Kling confirms they are live.
Does Kling 4.0 generate sound?
Yes. Kling 4.0 generates audio with the video, and launch coverage says it moved from single-channel to two-channel stereo. It also does lip-synced dialogue, which Kling says covers more than ten languages. The lip sync quality claims haven't been independently benchmarked, so test your language before a big project.
Kling 4.0 vs Seedance 2.5: which is better?
On paper they are close: both make up to 30 seconds with sound, and both reach 1080p in published docs. Kling 4.0 offers more control, with up to 15 references, 10 keyframes and an edit mode. There are no independent side-by-side benchmarks yet, so the honest answer is to run the same prompt through both.
Is there a Kling 4.0 API?
Kuaishou said enterprise and developer API access would "soon be made available," and reports from October 21 say Kling began rolling it out. Kling's public developer price table still listed only Kling 3.0-series models when we checked on November 2. Check availability on your own account before building on it.
Is Sora still available?
No. OpenAI announced on March 24, 2026 that it would discontinue Sora. The Sora app and web experience closed on April 26, 2026, and the Sora API was discontinued on September 24, 2026. OpenAI has not announced a replacement video model, so Kling, Veo, Wan and Seedance are the main options left.
Sources
- Kuaishou: Kling AI Officially Launches Next-Generation Kling 4.0 (press release, Oct 16, 2026)
- Let's Data Science: Kling AI opens Kling 4.0 to all users
- Pandaily on X: Kling 4.0 opens to all users
- SandBase: Kling 4.0 launch, what shipped and what didn't (Oct 23 tests)
- Grainzilla: Kling 4.0 Flash early access
- explainx: Kling 4.0 launch, 30-second video, stereo audio, Flash, MCP/CLI
- Wikipedia: Kling AI
- Framesurfer: Kling AI pricing (updated Oct 13, 2026)
- Kling AI developer pricing table
- Google Cloud: Veo 3.1 model reference
- Google Cloud: Extend Veo videos
- Google Gemini API: video generation docs (Veo 3.1 Preview shutdown)
- OpenAI Help Center: What to know about the Sora discontinuation
- Wikipedia: Sora (text-to-video model)
- TechNode: Alibaba launches Wan3.0
- Dataconomy: Alibaba launches Wan3.0, a 30-second AI video model
- Segmind: Wan 3.0 Video model specs
- ComfyUI docs: ByteDance Seedance text-to-video node (2.0, 2.5 specs)
Ask the question this article didn’t answer.
Ask anything about kling 4.0. Cleus opens this article, searches the live web and gives you a straight answer, with sources.
