Kling 4.0 and the AI Video Generation Race, Explained

Kling 4.0 Bets on Longer, Directed Clips

A few years ago, a 15-second ad for a local bakery meant a camera operator, a lighting kit, half a day of shooting and a few days of editing. On 28 September, Kuaishou announced Kling 4.0, an AI video generation model it says can produce up to 30 seconds in a single pass, follow up to ten keyframes, and handle long takes and continuous narrative better than before. The full version is promised for October. As of this week only a lighter Kling 4.0 Flash is open, as a trial for top-tier subscribers, according to Huxiu and ifeng.

The specs point to a bigger shift in the AI video generation race. For two years the contest was about short clips that rarely held together beyond a few seconds. Kling 4.0 and its rivals now compete on whether a creator can direct a short film from start to finish, keeping the same face and the same product on screen long enough to tell a story. That shift matters whether you buy ads, make short videos or just watch them on the train.

We have not tested Kling 4.0 Flash, which is limited to top-tier annual members; what follows on Kling relies on the announcement and on public reporting.

How the AI Video Generation Race Got Here

The first commercial video models were built to make a few seconds of plausible motion. Faces melted, hands multiplied and the camera wandered. The next generation stabilised the subject and learned camera language, so a prompt for a slow dolly push produced something close to one. Each step bought a little more time before the image drifted, which is why clip length became the headline number.

A marketer’s real problems are about control. They need the same bottle with the same label in shot one and shot five, a voice that matches the lips, and a result that does not take fifty attempts. Current models answer with inputs: reference images and videos that pin down a character or product, keyframes that fix where the story has to be at a given moment, and continuation that picks up where the last clip ended.

Kling 4.0 is a clear example. Kling says it accepts up to 15 references (up to ten images and five videos, covering up to seven subjects) and up to ten keyframes, according to ifanr and ifeng, and ifeng reports a prompt limit of 8,000 tokens. 4K and 1080p output in 10-bit HDR, and extending one video to two minutes, were announced as coming soon.

None of this has been tested independently yet. What remains hard for every model is consistency across many shots, believable lip sync, brand safety, and the cost of the attempts it takes to get one usable take.

Which AI Video Tools Are Worth Knowing in 2026?

This is a representative snapshot, not a ranked review, and it leaves out several capable Chinese models such as Hailuo, Vidu and Wan. Every capability below is the company’s own claim, and release cycles in this market run in months.

Kuaishou reported Kling revenue of more than RMB850 million in the second quarter of 2026, up over 200% from a year earlier, in its results release. In July it disclosed in a Hong Kong exchange filing that outside investors, including Tencent-controlled entities, are subscribing for new shares in its Kling subsidiary, cutting its own stake from 100% to about 68% if fully subscribed. Bloomberg reports Kling has picked banks for a Hong Kong listing; that is market talk until a filing appears.

Competition is close. ByteDance’s Seedance 2.5 already offers “up to 30 seconds per generation, with multi-round extensions”, and OpenAI shut down its Sora app and API this year. For a small brand, the practical difference is where the tool lives. Seedance reaches advertisers through TikTok’s Symphony creative tools, Google’s models sit in the Gemini API, and Kling sells direct subscriptions and an API. We walk through a full production process in our step-by-step guide to AI marketing videos.

What AI Video Changes in Digital Marketing

Creative: more versions, and a risk of sameness

The obvious gain is volume. A brand that used to run one hero video can now test a version for each audience: a different city in the background, a different opening line, a product in five colours. Localising a spot for another market becomes a regeneration, with no reshoot. The risk is that everyone uses the same models with similar prompts and feeds fill with ads that look alike. If every brand works that way, polish alone will not make an ad stand out.

Cost: where the money moves

A traditional shoot spends on people, places and calendars: crew, talent, location, permits, and the reshoot when the weather turns. A generative workflow spends on credits or subscriptions, on someone who can prepare references and write precise prompts, on a human editor who fixes the shots the model got almost right, and on brand and legal review of every version.

Our read is that the bill for a single ad falls for many formats, while the total creative budget may not, because teams spend the savings on more variants and more review.

Advertisers: from one film to a library

The platforms are already set up for this. Meta said in January that the combined annual revenue run-rate of ads using its video generation tools “hit $10 billion” in the fourth quarter of 2025, according to its prepared remarks. The IAB found in July that “nearly two-in-three buyers now use GenAI for digital video creative”.

The unit of work shifts from a campaign film to an asset library refreshed every week. That changes approvals (who signs off on 40 versions?), compliance (each version is a new claim to check) and attribution, because the winning variant may be one nobody on the team would have picked.

In our view, small advertisers gain the most, since a coherent 30-second story was previously out of their reach. Platforms gain too, because more creative gives their auction systems more to optimise, which tends to raise what advertisers spend. Agencies face the hardest adjustment: production margin shrinks, and their value moves to strategy, taste and running the testing. We described the same pattern in search ads in our piece on Google’s AI Max auto-upgrade.

How AI Video Is Changing Short Video and Short Dramas

China’s micro-drama industry shows where this goes fastest. Its broadcasting regulator said the market passed RMB100 billion in 2025 with more than 800 million viewers, and that of 430,000 micro-dramas released between January and August 2026, “AI-generated titles” made up “more than 90 percent”, as Global Times reported. The regulator did not say how much of each title was AI-made. At that volume, the constraint shifts from making content to getting it watched.

Our expectation is faster scrolling, and a harder time telling what was filmed from what was generated. That may create a premium for things a model cannot easily fake: a recognisable host, a live moment, a real place people can visit.

Platforms have a split incentive. Cheap AI content fills the feed, but a feed of near-identical dramas loses viewers, so ranking systems and labelling rules become the quality control. China’s labelling measures have required visible labels and metadata marks since September 2025. Creators, meanwhile, can produce more while each piece earns less, which favours those who build an audience over those who only sell footage.

Can AI Video Make Ads and Entertainment Better?

Lower cost also opens new formats. A family restaurant can afford a short narrative ad with a beginning and an end, where it used to run a slideshow of dishes. A campaign can be produced in a regional language, with subtitled and dubbed versions a full shoot would never have budgeted. A short-drama studio can try a genre that might fail, because a failed pilot costs days. Kuaishou says Kling-made ads won one Silver and two Bronze Lions at Cannes this year.

People do not disappear from this picture. Actors, directors and writers move toward the parts a model handles worst: performance that carries emotion, a world with its own rules, the judgement of what to cut. None of this improves the industry by default. It needs clear disclosure of AI content (the EU AI Act’s transparency duties have applied since 2 August), protection against impersonation, consent for likeness and voice, settled rights to training material, and platforms willing to demote spam.

What Kling 4.0 Means for Advertisers and Creators

Kling 4.0 is one milestone in a race that will produce another next quarter. The useful judgement to carry forward is that controllable length plus a real workflow is the commercial turning point, and that whether the result is better depends on guardrails and taste.

For advertisers:

• Build a reference kit before you generate anything: product shots from every angle, logo files, brand colours, and the faces you have rights to use. Models are only as consistent as their references.
• Budget for review. Every variant is a new claim.
• Test variants against each other in the ad platform, and let the results pick the winner.

For creators:

• Learn keyframes and references; they now do much of what camera direction used to.
• Label AI work clearly. China requires labels, the EU requires disclosure of deepfakes, and most platforms have their own rules; clear labels also keep audiences’ trust.
• Invest in what a model cannot copy: your voice, your audience and your point of view. If you want help turning AI video into a testing programme, that is part of our digital marketing work.

Further reading, by search term: Kling 4.0 release date; Kling vs Seedance 2.5; Gemini Omni video generation; IAB 2026 digital video ad spend report; EU AI Act Article 50 transparency.

FAQ: Kling 4.0 and AI Video

When is Kling 4.0 released?

Kuaishou announced Kling 4.0 on 28 September 2026 and said the full version would launch in October. A lighter model, Kling 4.0 Flash, opened the same day as a limited trial for Kling’s top-tier annual members. As of 12 October 2026 the full version had not been released.

What can Kling 4.0 do?

According to Kling’s announcement as reported by Chinese media, Kling 4.0 generates up to 30 seconds of video in a single pass (up from 15), accepts up to 15 reference inputs including 10 images and 5 videos, and lets creators set up to 10 keyframes. 4K 10-bit HDR output and extending a video to two minutes were announced as coming soon. These are company claims that have not been independently tested.

How does Kling compare with Seedance, Veo and Runway?

ByteDance’s Seedance 2.5, released on 31 July 2026, also generates up to 30 seconds per pass with multi-round extensions and joint audio. Google now offers Gemini Omni alongside Veo 3.1, which has native audio. Runway’s Gen-4.5 added native audio and longer multi-shot video, and Luma’s Ray3.2 makes 20-second 1080p clips with HDR. OpenAI discontinued Sora in 2026. No independent benchmark settles which AI video generator is best; the right choice depends on the workflow.

Will AI video make ads cheaper?

It moves the cost rather than removing it. Shooting days, crews and locations shrink for many formats, while new costs appear in credits, prompt and reference preparation, human editing, brand and legal review, and testing many variants. The largest change is the volume of versions an advertiser can afford to test.

Do AI-generated videos have to be labelled?

Often, yes. China’s labelling measures for AI-generated content took effect on 1 September 2025 and require visible labels plus metadata marks. The EU AI Act’s transparency duties for deepfakes and AI-generated content have applied since 2 August 2026. Platforms add their own disclosure rules on top.