Kling 4.0 review: thirty-second product videos, if someone checks the product

The short version

Kling 4.0, the new video model from Kuaishou's Kling AI, became fully available on October 15, 2026. It makes clips of up to thirty seconds in one go, with sound, and it lets you steer them with up to fifteen reference images and videos. For a store that needs short product videos and social clips before the holidays, it's one of the more capable options out right now.

The condition is the same one we'd attach to any generative tool used for products: someone who knows the product has to look at every clip before it runs. A thirty-second clip is thirty seconds of chances for the product to change shape. And two of its headline features, recreating a video that already works and borrowing a voice style, need a rights conversation before anyone uses them on a paid ad.

Who it's for: ecommerce and brand teams making social clips, product demos and ad variations, with a person who can sign off. Who should skip it: anyone hoping to publish straight from the generator.

What Kling says is new

From Kling's release notes, user guide and launch post:

• Length. Clips from 3 to 30 seconds in a single generation, at 720p, 1080p or 4K, in 21:9, 16:9, 1:1 or 9:16.
• Sound. Two-channel stereo audio and, Kling says, more accurate lip sync.
• References. Up to 15 reference assets per generation, including up to 10 images and up to 5 videos totaling 30 seconds, plus up to 10 keyframes to pin down specific moments.
• Longer prompts. Up to 8,000 tokens.
• Text. Kling claims clear, stable rendering of multilingual text and logos, even with camera movement.
• Recreation. A feature Kling calls Creative Recreation references the pacing, shot composition and look of an existing video while replacing the products and characters.

Two items were still marked coming soon at launch: 10-bit HDR output, and API access for enterprises and developers. Access at launch was through kling.ai, plus MCP and a command-line tool.

Green flags

Thirty seconds is a real ad length. Most product clips people actually run are well under that. Getting a whole demo in one generation, rather than stitching short pieces, removes a lot of fiddly editing.

References and keyframes are the right controls for products. The biggest risk with product video is drift: the bag gains a buckle, the bottle changes shape. Being able to feed in up to ten product photos and pin key frames is the kind of control that makes fidelity achievable. It doesn't guarantee it.

Sound in the same pass. For social clips, generating audio alongside the picture saves a step, provided the audio is checked too.

Pricing is published per second. Kling's developer price table lists Kling 4.0 at $0.157 a second at 720p and $0.235 at 1080p when you don't supply an input video. That makes a budget easy to estimate before you start.

Agents can drive it. MCP and a command-line tool mean it can sit inside a production workflow rather than only a web page.

Red flags

Fidelity is your job. Kling's material is about realism, motion and camera work. None of that is the same as showing your actual product correctly for thirty seconds. Plan for someone to check shape, color, material, logo and scale in every frame that matters, and to reject freely.

Recreating a trending video is a rights question. Copying another video's pacing and shot composition while swapping in your product is pitched for social and product advertising. It may also land you close to someone else's creative work. Get a clear internal rule on what can be referenced before your team discovers the feature.

Voices and faces need consent. Voice-style references and better lip sync make it easy to produce a convincing person saying something. Use only voices and faces you have permission to use, and say so in your records.

Some things aren't there yet. HDR was coming soon at launch, and Kling's price table, as of October 22, lists no 4K rate for 4.0. If you need 4K masters through the API, confirm what your account can actually get.

Text on the product still needs reading. Kling says text and logos stay clear and stable. That's a claim worth checking every time, because a label that reads almost right, an ingredient list with one word changed, or a size printed wrong on the box is the kind of error a shopper screenshots.

Retries are where the money goes. A clean thirty-second clip at 1080p costs about $7 on the published rate. If it takes six attempts to get the product right, it costs six times that.

Price, and the obvious alternative

The obvious alternative is Kling's own previous model. On the same price table, Kling 3.0 with native audio is $0.126 a second at 720p and $0.168 at 1080p. A thirty-second 1080p clip works out at about $7.05 on 4.0 and about $5.04 on 3.0 with audio, before retries. If your clips are short and simple, 3.0 may be enough. If you need the length, the extra references or the keyframes, 4.0 earns the difference.

Supplying an input video changes the billing. Kling bills the combined duration of input and output in some cases, so read the rules before you budget for recreation or editing work.

Subscription plans on kling.ai use credits instead. Check the current plan prices on Kling's site rather than a blog post, since they've moved with each launch.

Our verdict for a store owner

Worth a trial project before the holidays, with three rules in place first. One person owns product sign-off for every clip. Nothing based on another brand's video or anyone's voice runs without a rights check. And you set a budget per finished clip, including retries, before anyone starts generating.

We've built this kind of review into AI imagery before. TWIMCO built Nines Style's AI generation suite, which turns product photos into images of models wearing the products and puts image review inside the workflow, because a convincing picture of the wrong product is worse than none. Our guide to reviewing AI-generated product imagery applies to video frame by frame.

If you want generated product video inside your own catalog and review workflow, rather than in someone's downloads folder, that's the AI applications and agents work we do. Talk to us.