Pricing+7% bonus

What Is Kling 4.0? Preview Leaks, Release Status, and What You Can Use Now

Is Kling 4.0 available yet? Explore the Preview and Flash leaks, what remains unconfirmed, how it compares with Kling 3.0, and which models you can use now.

What Is Kling 4.0? Preview Leaks, Release Status, and What You Can Use Now

Status checked: September 28, 2026. This article will need an update when Kling publishes a 4.0 announcement or API documentation.

The first question about Kling 4.0 is surprisingly practical: can you make a video with it today? I cannot point to an official public Kling 4.0 release or a documented 4.0 API model ID yet. What people are calling “Kling 4.0” is a possible next generation of Kuaishou's Kling video models, discussed after screenshots and a feature list circulated online. Kling's public site still presents the 3.0 series as its current lineup.

That distinction matters if you are planning a product video or wiring an API into a production workflow. A setting seen in a preview screen is a clue about development. It is not a promise about the model you can buy, call, or compare today.

Table of contents

Is Kling 4.0 Released? What Happened in September 2026

On September 24, an X post describing a Dev Mode leak named VIDEO 4.0 Preview, VIDEO 4.0 Flash Preview, and a separate Image 4.0. The accompanying image appears to show model names in a creator interface. Neither the post nor the image establishes a general release, an official API endpoint, or final specifications.

On September 27, Kling's official account said it had news. The post did not say that the news was Kling 4.0. Reading the two posts together makes a 4.0 announcement seem plausible; it does not supply a release date. That is my inference, not a confirmed rollout schedule.

For now, the useful answer to “What is Kling 4.0?” is: a prospective Kling video generation, with Preview and Flash names reported in a leak, whose public features, price, and availability have not been confirmed by Kuaishou.

What Do the Kling 4.0 Preview Leaks Actually Claim?

Most of the eye-catching numbers trace back to the same September 24 post. Repetition across model directories does not turn one unverified report into several independent confirmations.

Reported detail What we can establish now
Up to 30 seconds per video generation Claimed in the post. No public 4.0 documentation or repeatable output test confirms the limit.
Up to 1080p video Reported alongside the video claims. This is not evidence for 30-second 4K video.
15 Omni Reference elements and up to 3 voice references Numbers from the post. It does not establish that all 15 can be images, or which combinations will work in each mode.
Video extension to 60 seconds A reported extension workflow, distinct from generating a 60-second clip in one pass.
Experimental long-video mode up to 120 seconds An early claim about a separate mode, not a demonstrated two-minute production result.
Flash Preview A name in the circulated material. There are no confirmed speed, quality, or price figures for a Flash-versus-standard comparison.

These are reported settings, not Kling 4.0 features I have tested. The limitations may change, and access could differ by plan or task. There is also a naming trap: the same post discusses Image 4.0 generating multiple images, with possible 4K or 8K image settings. Those image rumors say nothing about the resolution of a 4.0 video clip.

The reference count is worth watching for creators who reuse a character or product across shots. But a larger upload allowance alone would not prove better consistency. The result has to preserve the subject after a camera change, a new background, and an action involving other objects.

Kling 4.0 vs Kling 3.0: What Can We Compare Today?

Kuaishou announced the Kling 3.0 family on February 5, 2026. Its announcement describes clips up to 15 seconds, native audio, reference-guided creation, and multi-shot storytelling. On April 23, Kling added native 4K video output to its 3.0 series. Different 3.0 models and tasks can expose different settings, so check the exact route before treating a family-level feature as universal.

Question Kling 3.0 Kling 4.0
Public model family Officially announced Preview names reported; official launch not established
Video duration Up to 15 seconds in Kuaishou's 3.0 announcement Up to 30 seconds claimed, unverified
Native 4K video Officially added to the 3.0 series Not established by the leak
Audio and multi-shot controls Described in the official 3.0 announcement Final 4.0 modes and limits unknown
Published API specification and price Check the current model and provider pages No verified public 4.0 specification or price in the sources reviewed

Here is the correction that matters most: Kling 4K is a resolution feature of the existing 3.0 generation; “Kling 4.0” would be a version name. Combining the leaked “30 seconds” with 3.0's confirmed “4K” creates a specification nobody has established. A 1080p setting in a leaked interface also cannot prove that a later 4.0 release will never offer 4K. Both claims go further than the evidence.

Need native 4K now? See the currently listed Kling 3 Omni 4K model on GPT Proto. Check that model's task and pricing panel before starting a batch; it is a 3.0 route.

Kling 4.0 vs Seedance 2.5: What Would Settle the Comparison?

There is no honest head-to-head verdict before Kling 4.0 can be run under the same conditions. A longer maximum duration would help a creator fit more of a scene into one job, but it might also make a failed take more expensive to redo. More references could improve direction, or simply add conflicting instructions. Both are hypotheses.

I would test the models with the same short briefs, aspect ratio, reference assets, and output target:

Test scene Check the delivered clip for
A person walks into a café, sits, picks up a cup, then speaks across two angles Face, clothing, hand-to-cup contact, continuity, and lip sync
A product rotates under a changing light, then appears in a second setting Label legibility, shape, color, and logo consistency
A shot requiring one small correction Whether changing the requested detail also changes the rest of the scene

Record render time, failed attempts, and cost per accepted clip, not just the price of the first generation. One creator discussing a 30–60-second animation series describes how a character and visual style can drift between individually convincing shots. That is a real production question to test. It is not evidence that Kling 4.0 has solved it.

Want a live comparison baseline? Seedance 2.5 is listed on GPT Proto. Use the capabilities shown for its specific API task rather than assuming every feature of the wider model family is enabled on that route.

Can You Use Kling 4.0 Through an API? What to Try Now

An interface screenshot does not define an API. Before integrating Kling 4.0, a developer needs a published model ID, input schema, supported tasks, output constraints, error behavior, and billing unit. We have not established those for a public 4.0 endpoint, nor a date when GPT Proto will list one. A third-party page labeled “Kling 4.0 API” is not proof that it calls a released Kuaishou 4.0 model.

You can work with a currently listed Kling model instead. For example, the GPT Proto Kling 3 Omni 4K model page shows this text-to-video request. It is a Kling 3.0 example, not a preview 4.0 call:

export GPTPROTO_API_KEY="YOUR_API_KEY"

curl --request POST \
  'https://gptproto.com/api/v3/kling/kling-v3-omni-4k/text-to-video' \
  --header "Authorization: Bearer $GPTPROTO_API_KEY" \
  --header 'Content-Type: application/json' \
  --data '{
    "prompt": "One unbranded ceramic coffee cup on a wooden table. The camera slowly moves closer as morning light crosses the surface.",
    "sound": false,
    "aspect_ratio": "16:9",
    "duration": 5
  }'

Set your own key before running it; a successful submission may use your balance. Follow the Check Status instructions on the same model page to retrieve the completed video. The request structure is taken from that live model page, but I have not executed a paid generation with a GPT Proto key for this article. Verify the response and billing in your own account before building on it.

Ready to try a model that is listed today? Browse GPT Proto's Kling models, choose the task you need, and use the quick start attached to that model page. A future 4.0 route should be added here only after its model ID and behavior are verified.

Frequently Asked Questions

When is the Kling 4.0 release date?

Kuaishou has not provided a release date in the official material checked on September 28, 2026. Its September 27 teaser did not name Kling 4.0. Dates inferred from that teaser or the preview screenshot are guesses.

Is Kling 4.0 Flash faster or cheaper?

The circulated material names a Flash Preview variant, but it does not supply verified generation-time or pricing data. “Flash” is a label, not a benchmark or a price quote.

Can Kling 4.0 generate a 30-second video in 4K?

That combination is unverified. The leak claims up to 30 seconds and mentions up to 1080p for video. Native 4K video is officially documented for Kling 3.0; a separate rumor about 4K or 8K concerns Image 4.0, not 4.0 video.

Can I put 15 references in one Kling 4.0 video?

One unverified post claims an Omni Reference limit of 15 elements. It does not establish the accepted mix of images, videos, and other elements for a public 4.0 video request. Wait for the final input specification before designing an automated workflow around that number.

Related Articles

More Blogs
6 Best LLM API Providers in 2026: Multi-Model Platforms Compared

6 Best LLM API Providers in 2026: Multi-Model Platforms Compared

Choosing an LLM API provider is no longer the same as choosing a model. The same open-weight model can be available from several platforms, yet the real service you receive may differ in latency, throughput, context limits, tool calling, caching, error behavior, and price. The lowest listed token price can cost more in production if cache hits are unreliable or retries are frequent. An “OpenAI-compatible” endpoint may also accept basic chat requests while rejecting fields your application needs. We compared six multi-model LLM API providers across aggregators, managed cloud platforms, and inference specialists. First-party APIs such as OpenAI and Anthropic remain useful baselines, but they do not offer the same cross-vendor access. One Key for Your Team

Tiffany Layne | 2026-09-21

How to Make an AI Factory Video: 3 Easy Steps for Beginners

How to Make an AI Factory Video: 3 Easy Steps for Beginners

A factory video can show more than machines moving. In ten seconds, it can introduce a worker, show the material changing, and end on the product that gives those movements a purpose. The hard part is keeping the production sequence believable while making the shots feel like one film. This guide uses Morrow Cocoa Works , a fictional dark-chocolate manufacturer, as a working example. You will plan a 3 × 3 storyboard, write a video prompt with nine timed shots, and finish the brand text in an editor. The prompts are ready to adapt; the fictional chocolate film has not been rendered or tested for this article. If you are making a video for a real company, substitute its approved product photographs, facility details, and production steps before generating anything. The short version of the workflow is: verify the process, make a storyboard sheet, generate video from approved references, then inspect and edit the result. A storyboard is a visual plan, not nine automatically editable video clips. Create AI Factory Video

Tiffany Layne | 2026-09-20

How to Make an AI Cat Dance Video With Your Own Cat

How to Make an AI Cat Dance Video With Your Own Cat

To make an AI cat dance video, you need a clear image that tells the model what your cat looks like and a second input that tells it how to move. That movement can come from a dance reference video or from a written prompt. The reference-video route is the better choice when you want your cat to copy a particular hand dance, K-pop routine, side step, or meme choreography. A prompt is enough when you only want a general result such as “a tabby cat doing a playful hip-hop dance.” It is quicker, but the model decides more of the choreography for you. This guide covers both workflows with Wan 3.0. If you already have a cat photo and a dance clip, you can skip most of the setup and use the GPTProto AI Cat Dance Video Generator . Create Cat Dancing Video

Tiffany Layne | 2026-09-17

5 Best Affordable AI Video APIs in 2026: Pricing, Ecommerce, and Short Drama

5 Best Affordable AI Video APIs in 2026: Pricing, Ecommerce, and Short Drama

Vidu Q3 Turbo is the best affordable AI video API for most developers in 2026. A five-second 720p clip costs about $0.24, while 1080p costs $0.056 per generated second. Seedance 2.0 Mini is better for cheap drafts, Hailuo 2.3 Standard for fixed six-second action clips, Kling 3.0 Standard for dialogue, and Wan 3.0 for longer multi-shot stories. Those winners change when you add resolution, minimum clip length, audio, and failed attempts. This comparison looks beyond the lowest advertised rate to estimate what each API costs for the shot you can actually use. Price note: GPTProto prices in this guide were checked on September 15, 2026. Video API rates and available settings can change, so confirm the live model page before budgeting a production run. Try Vidu Q3 Turbo

Tiffany Layne | 2026-09-16