Kling 4.0 Is Out: 9 Things to Know Before the API

The waiting part is over. The building part hasn't started. That's the strange spot every video team is in this week: the model everyone spent a month speculating about is finally out, you can watch people post 30-second clips from it, and you still can't put it behind a button in your own product.

Kling 4.0 went fully live on October 15, 2026. Kling announced it on X with a "What's New, Starting Today" post, and its official global launch release followed on October 16. The release confirms some things we'd only seen in early access, quietly moves a couple of headline features back to "soon", and adds details nobody was talking about before launch. Here are the nine that matter if you make video for a living, plus what to do on each::labs while the API catches up.

1. Kling 4.0 Shipped, but Not as an API

This is the one to get straight first. Kling 4.0 is fully available on the Kling AI website, through what Kling calls a refreshed creation interface. Kling also opened access through MCP (Model Context Protocol) and a command-line interface from day one, so you can drive it from an agent or a terminal instead of the web editor. Launch-day coverage in RuntimeWire, citing Kling's X post, sums up the split well: everything above was live on October 15, while API access for enterprise customers and developers sat under "coming soon".

The October 16 release repeats the same line. API access for enterprises and developers is coming soon, with no date attached. So if you've seen a claim that Kling 4.0 is "available everywhere", read it as "available to people clicking around Kling's own site". Production pipelines that call a model programmatically are still waiting, and so are we. It isn't on each::labs yet, and we won't pretend otherwise.

2. Two of the Headline Features Are Still "Soon"

Early-access chatter treated 10-bit HDR and the 21:9 ultrawide format as done deals. The official launch release is more careful. It says Kling 4.0 will "soon" support 10-bit HDR output at 4K and 1080p, and lists the 21:9 format in the same breath. Neither is described as live today.

That matters more than it sounds. Kling pitches HDR as the feature for backlit, night and neon scenes, where more detail survives in the highlights and shadows and you get more room in the color grade. If your brief is a moody cinematic spot, plan it so it still works in standard dynamic range, and treat HDR as an upgrade you get later rather than the reason the concept works. Our earlier roundup of Kling 4.0 rumors flagged 21:9 as something early testers showed. The launch release is now the source to trust on what's actually shipping.

3. Thirty Seconds Is the Take. Two Minutes Is the Ceiling.

The confirmed number hasn't moved: Kling 4.0 generates natively from 3 to 30 seconds in a single generation. What the launch adds is the extension story in plain terms. You can upload an existing video and extend it to a total of up to two minutes.

Don't read that as "two-minute generations". It's 30 seconds you direct as one piece, then a continuation workflow on top. The practical consequence is that your first 30 seconds become the anchor for everything after it, so the opening clip deserves more care than it used to. If the character, the light and the camera language are wrong at second 10, extending to second 110 just gives you more of the wrong thing. We covered how to plan a half-minute in beats in the Kling 4.0 prompting guide, and that planning now pays off twice.

4. White-Model and Depth-Map References Are the Sleeper Feature

This one barely came up before launch, and it might be the most useful thing in the release. Kling 4.0 accepts white-model and depth-map reference videos to guide character performance, action, camera movement and pacing.

A white model is the plain, untextured grey figure you get out of a 3D blockout. A depth map is a grayscale video where brightness stands for distance from the camera. Both carry motion and space without carrying a look. That's the point. You can block a scene in rough 3D, or pull depth from footage you already shot, and hand Kling 4.0 the timing and camera path while the prompt and your other references decide what everything looks like.

If that sounds familiar, it's because it's the same instinct behind Kling V3 Pro Motion Control, where you copy a move from a reference clip onto your character. The difference is scope. Motion control copies a performance. A depth or white-model pass can describe the whole shot, camera included. Our step-by-step motion control guide is still the best way to learn the habit today.

5. References Now Include Sound

Kling 4.0's Omni Reference system takes up to 15 multimodal references, and the launch release spells out that this means images, video and audio. Kling's own keyframes guide describes the same pool as up to 15 combined assets per task, including subjects and voice references.

Fifteen is a lot of inputs, and more isn't automatically better. Every reference you add is one more thing the model has to reconcile. The best results we've seen from reference-heavy models come from giving each input exactly one job: this image is the face, this clip is the walk, this audio is the voice. You can practice that discipline right now with Kling O3 Pro Reference to Video, which is built around the same idea.

6. Ten Keyframes Turn a Prompt Into a Shot List

Up to 10 keyframes in a single generation is confirmed. Kling's guide is clear about the division of labour: keyframes define the important visual states, and the prompt explains how the video moves between them. Its own 30-second example drops keyframes at 0, 6, 12, 20 and 28 seconds, which is a useful reminder that you don't need all ten.

The guide also warns about something that will bite people in the first week. Keyframes and references shouldn't contradict each other. If your reference image puts a character in a red coat and keyframe four shows a blue one, you've asked for a costume change whether you meant to or not. Pick one source of truth per detail.

7. You Get 8,000 Tokens of Prompt. Use Them for Structure.

Prompt input on Kling 4.0 goes up to 8,000 tokens. That's room for an actual treatment, not a caption. The trap is treating it as permission to pile on adjectives. A longer prompt helps when it adds order (what happens first, what the camera does next, where the cut lands) and hurts when it just repeats the mood in six different ways.

Kling's release also says camera work got steadier: push-ins, pull-outs, pans, tracking shots and orbits are described as more natural and stable, with smoother transitions and scene changes. Spend your tokens describing camera moves in the order they happen. That's where a long prompt actually earns its length.

8. It Speaks More Languages, and More Accents

The launch release lists improved support for English, Japanese, Korean, Spanish, Portuguese, German, French and Hindi, "and more". It also names accents, not just languages: American English, British English and Indian English.

Accent control is a small line in a press release and a big deal for anyone localizing ads. A spot for London and a spot for Mumbai have always needed more than a translated script. If you're planning multilingual campaigns, write your dialogue lines per market now, and keep the visual setup identical across versions so the only variable is the voice. In the meantime, Kling Voice Create on each::labs is where you can start building those voice assets.

9. Real Productions Got There First

Kling chose to launch with filmmakers rather than benchmarks. The release quotes John Adams and Diane Shorthouse, co-founders of Evolutionary Films, who used Kling 4.0 on the animated feature Minibots. Their take: "It feels increasingly like directing a film rather than prompting an AI." They single out the Omni Reference workflow, camera work and multi-character consistency.

Sung-Soo Kim, director of 12.12: The Day, used AI for the first time on the short film Worm Train, recreating a subway scene with a preview build in September. His line was about physics: "I was able to see the realistic physical interaction and sense of weight." These are partner testimonials, not independent tests, so weigh them that way. But it tells you who Kling built Kling 4.0 for. Not people making one viral clip. People making things with a running time.

How to Get Ready for Kling 4.0 on each::labs

Kling 4.0 is coming to each::labs, and the API is the gate. Until it opens, the smart move is to build the workflow so that switching models is a one-line change, not a rebuild.

Start with the Kling models that are live today. Kling V3 Pro Text to Video and Kling V3 4K Text to Video cover most single-shot work, and Kling O3 Pro Image to Video is the one to use when a keyframe-style starting image matters. Write your prompts in beats, keep your reference library organized by job (face, wardrobe, motion, voice), and store depth or blockout passes for shots you'll want to re-run.

Then wire it into a flow. With each::labs AI Flows you can chain image generation, video and voice into one pipeline today, and slot Kling 4.0 in as the video step the day it lands. Browse the wider text-to-video and image-to-video catalogs if you want a second model in the chain for comparison. For the full background on what Kling promised before launch, our original Kling 4.0 explainer and the Kling 4.0 use cases post are still accurate on the core specs.

Frequently Asked Questions

Is Kling 4.0 available through an API yet?

Not publicly. Kling's launch materials list API access for enterprises and developers as coming soon, with no date. Right now you can use Kling 4.0 on the Kling AI website and through Kling's MCP and CLI access.

When will Kling 4.0 be on each::labs?

As soon as Kling opens developer API access. We'll publish the model page then, and the links in this post will point straight to it.

Does Kling 4.0 support 10-bit HDR and 21:9?

Both are announced, and both are marked "soon" in the official launch release. HDR is planned at 4K and 1080p.

How long can a Kling 4.0 video be?

A single generation runs from 3 to 30 seconds. By uploading an existing video and extending it, you can reach up to two minutes in total.

What's new in Kling 4.0 compared with Kling 3.0?

The biggest jumps are length, control and inputs. Thirty-second native clips, up to 10 keyframes, up to 15 image, video and audio references, white-model and depth-map reference videos, and 8,000-token prompts. Kling V3 is still the one you can run on each::labs today.