Kling 4.0: what it is, what changed, and what you can use today

Kling 4.0 is announced, and only Flash is out so far

Kling AI, Kuaishou’s video model business, introduced Kling 4.0 at the end of September 2026. Its own guide, published on 30 September, says two things about timing: the full model “is scheduled to launch in October 2026”, and early access to the new creation experience “began rolling out to a limited group of users on September 28”. No launch day has been given.

What that limited group has is Kling 4.0 Flash, a lighter, faster member of the 4.0 family. Kling’s feature page (read on 4 October 2026) describes it as available “in a limited rollout”, and suggests signing in to see whether your account is eligible. If you can’t see it, nothing is wrong with your account: most people can’t yet.

Most of the excitement you will see about Kling 4.0, the 30-second clips, the keyframes and the 4K HDR, describes the full model. Flash is a different, smaller thing, so it is worth knowing which is which before you plan a project around either.

Flash and the full model side by side

All of this comes from Kling’s own Kling 4.0 feature page, as of 4 October 2026.

Two of the full model’s features are marked as coming even after launch: Kling’s guide lists 10-bit HDR at 1080p and 4K as “coming soon”, and its feature page lists video extension, to a total of 2 minutes, as coming soon too.

So Flash is good for trying the new model’s motion, sound and references on short pieces now. If your project depends on long takes, keyframes or 4K delivery, it is waiting for the full model.

What changed from Kling 3.0

Kling published its own comparison on 30 September 2026, and it is clear and specific, which is to its credit. The changes that matter most to people making things:

• Twice the length. Kling 3.0 made 3 to 15 seconds; 4.0 makes 3 to 30. A whole story beat, or a short ad, can now be one generation rather than two clips stitched together.
• Keyframes. Kling 3.0 worked from a start frame and an end frame; 4.0 takes up to 10 keyframes. You decide what the frame looks like at key moments and the model fills in between, which is how you get a scene that lands where you planned instead of drifting.
• Far more references. Kling 3.0 took up to 7 images or subjects, or one video of up to 15 seconds plus 4 images or subjects. Kling 4.0 takes up to 15 references in one task: up to 10 images, up to 5 videos with a combined 30 seconds, up to 7 subjects, and voice references for a character’s voice.
• Better sound. Single-channel audio becomes two-channel stereo, and Kling says lip sync is more accurate.
• Colour. 8-bit SDR becomes 10-bit HDR at 1080p and 4K, once that part ships.
• Languages and text. Kling 3.0 handled Chinese, English, Japanese, Korean and Spanish; 4.0 adds Portuguese, German, French, Hindi and more. Kling also says on-screen text, logos and emoji hold up better while the camera moves, and that prompts can run to 8,000 tokens.

Kling’s guide also describes a redesigned creation page: one input box for every kind of asset, Grid and List views, editing on a timeline, and what it calls a Canvas and Agent experience for organising a project.

Kling hasn’t published side-by-side quality tests against other models, and we haven’t run any, so this page says nothing about whether 4.0 looks better than its rivals. Judge that on your own prompts once you have access.

Recreating and editing: the new way of working

Two features change how you might use Kling, not just how good the output is.

Recreation. Kling’s FAQ says 4.0 “can use an existing video as a reference for pacing, shot composition, and camera movement” while putting in your own characters and products. Its own example is taking a trending product ad and swapping in your product, characters and setting. That is useful for small brands. The usual care applies: borrowing a structure is one thing, copying someone’s distinctive work is another, and the brand or creator whose video you start from deserves credit where you can give it.

Editing. The full model can change subjects, backgrounds, styles, expressions, colours, materials and camera angles in footage you already have, from text and visual references, with up to five clips in one task. In Kling 3.0 that sat in the separate 3.0 Omni workflow; in 4.0 it is part of the main model.

Kling’s guide also gives some sensible advice for longer clips: break the piece into story beats, give each reference one clear job, write the prompt as direction (subject, action, setting, camera, lighting, pacing, sound), and review the whole sequence for continuity rather than judging single frames. With 30 seconds and 10 keyframes to plan, that is how a director would work anyway.

Where you can run Kling 4.0

As of 4 October 2026:

• Kling’s own app is the only place Flash is running, and it is where the full model is promised for October.
• Higgsfield offers a long list of Kling models, and its Kling help page is a good guide to them, but that page and its AI video page list Kling 3.0, 3.0 Omni, Kling o1 and earlier, with no 4.0 yet.
• fal says “Kling 4.0 is coming to fal soon”, and Morphic says 4.0 and 4.0 Flash will come to it “after Kuaishou’s release”. Neither has it yet.
• Kling’s developer API: we found no 4.0 model on Kling’s developer pages. If you build with the API, watch Kling’s developer platform after the full launch.

For what any of these cost, see each platform’s own pricing page; prices and credit costs change too often for a page like this.

Getting ready before the full launch

You don’t need access to start preparing, and the preparation is the same work that makes any video model behave.

Gather your references now. Fifteen references in one task is only useful if you have them ready: a clean set of pictures of each character from several angles, the product shots, the locations, and short clips whose movement you like. The more references a model takes, the more a tidy, reusable set is worth. In BosOn we call these elements: you make a character once and use them in any prompt, on whichever platform runs it (see elements on our home page).

Plan in beats and keyframes. Sketch the 30 seconds as beats, then decide which moments need a fixed frame. You can make those keyframes now with any picture model you already use.

Keep what you make in one place. Longer clips with more inputs are harder to reproduce, so when a 30-second take works, write down the prompt and which references and keyframes you used, somewhere you’ll find them again. BosOn keeps every picture and clip in one library, and makes new ones through your own Higgsfield account, so the Kling models you run there end up with everything else you’ve made. BosOn is Windows only and in early access; see the integrations page for what is ready today.

We will update this page when the full model launches and when it reaches other platforms.