Most AI video clips fall apart somewhere after the eighth second. The product changes shape, the logo melts, and the camera drifts into a shot nobody asked for. That is why most of us have been stitching together five-second clips and hoping the lighting matches.
Kling 4.0 changes the maths. It can generate a single clip of up to 30 seconds and lets you pin up to 10 keyframes along the way. But a longer clip only helps if you plan it like an editor, not like a prompt writer. Below is what changed, a "beat sheet" method for keyframes, a copy-paste prompt, and the places it still breaks.
What is Kling 4.0 and what changed?
Kling 4.0 is the fourth-generation AI video model from Kuaishou's Kling AI. It became available to all users on 15 October 2026 after early access began on 28 September. It generates 3 to 30 second clips at 720p, 1080p or 4K, accepts up to 10 keyframes plus start and end frames, and outputs two-channel stereo audio.
According to Kling's user guide as summarised by Wireflow's 18 October pricing report, the full model also adds:
- Omni Reference: up to 15 assets in one request, made up of 10 images, 5 video clips (30 seconds in total) and 7 elements such as characters or products.
- Long prompts: up to 8,000 tokens, enough for a full written script.
- On-screen text in nine or more languages, a 21:9 cinematic ratio, and editing of existing clips (up to 5 clips, 30 seconds in total).
There is also a faster Kling 4.0 Flash mode. It is capped at 20 seconds and 720p, takes a start frame only, and does not support multiple keyframes yet.
Two headline features are not here yet. 10-bit HDR and video extension up to two minutes are both marked "coming soon" in Kling's guide.
Why do keyframes matter more than a longer prompt?
Keyframes matter because they turn a 30-second clip from one long guess into a sequence of checkpoints. A keyframe is a reference image that the video must pass through at a set moment. With up to 10 of them, you decide what the product, person and setting look like at each beat, and the model only fills the motion in between.
A text prompt describes intent. A keyframe shows the answer. When you write "the cup slides into frame and the logo faces the camera," the model still has to invent the cup, the angle and the logo. When you give it an image of exactly that moment, there is nothing left to invent.
This is also how you protect brand details. Logos, packaging and price tags are where AI video drifts first. If the frame at second 24 shows your real packaging, the model has a fixed target to land on.
The trade-off is preparation time. You need still images for each beat before you generate anything. In practice, that is the same work as a storyboard, and it is far cheaper than rerunning 30-second generations until one happens to look right.
How do you plan a 30-second clip with a keyframe beat sheet?
Plan the clip as five to seven beats before you open Kling. For each beat, write the timestamp, what the viewer must see, the camera move, and any on-screen text. Then create one keyframe image per beat. Five well-spaced beats usually work better than using all 10 slots on a 30-second clip.
Here is a beat sheet for a 30-second coffee shop promo:
- 0s: wide shot of the shop front at morning light, door opening.
- 6s: barista pours latte art, close-up, slow push-in.
- 13s: customer takes the first sip at a window seat, smile.
- 21s: cup on the table, logo facing camera, steam visible.
- 27s to 30s: end card with the shop name and "Open 7am daily".
Three rules keep the beats clean. Leave at least four seconds between keyframes so the motion has room. Keep one subject change per beat. And make the last frame a simple end card, because text-heavy endings are easier to fix in an editor than in a regeneration.
For the keyframe images themselves, use an image model that holds characters and products steady. Our Nano Banana 2.1 consistency guide covers a reference-roster method that works well for this step.
What prompt should you pair with your keyframes?
Pair your keyframes with a short, labelled script that tells Kling what happens between them, not what is in them. Use five parts: FORMAT, BEATS (timestamp plus action), CAMERA, AUDIO and RULES. Describe motion and transitions only, because the keyframes already carry the look of each moment.
Try this prompt:
FORMAT: 30-second vertical 9:16 video, 1080p, warm natural light, realistic, one continuous take feel.
BEATS: 0s to 6s, the shop door opens and the camera glides inside (keyframe 1 to 2). 6s to 13s, the barista finishes a latte art pour, slow push-in on the cup (keyframe 2 to 3). 13s to 21s, cut to a customer at the window taking the first sip, gentle smile (keyframe 3 to 4). 21s to 27s, the cup rests on the table, steam rising, logo facing camera (keyframe 4 to 5). 27s to 30s, hold on the end card (keyframe 5).
CAMERA: smooth handheld gimbal, no fast pans, no zooms beyond a slow push-in.
AUDIO: soft cafe ambience, espresso machine hiss at 6s, light acoustic guitar, no voice-over.
RULES: keep the cup shape, logo and packaging exactly as in the keyframes. Do not add extra text, signs or logos. Do not change the barista's or customer's face or clothing between beats.
Notice what the prompt leaves out. It does not describe the barista's face or the cup design, because those live in the keyframes. Repeating them in text invites the model to reinterpret them.
The RULES line does most of the protective work. "Do not add extra text" alone removes a large share of garbled background signage, a common failure in AI video.
Before you spend a full 30-second generation, test the first two beats as a short clip in Flash mode or at 720p. If the motion between beats one and two looks wrong, the rest will too.
Where does Kling 4.0 still fall short?
Kling 4.0 still has real gaps: Kling has not published how many credits a 4.0 video costs, its API does not support 4.0 yet, the free plan does not allow commercial use, and Cantonese is not named among its listed languages. Plan around these before you promise a client a 30-second spot.
- Unknown credit cost. On 19 October, Wireflow found that Kling's plan and guide pages listed no per-video credit rate for 4.0 or Flash. Run one short test and note the credits used before you plan a batch.
- Plan prices. Wireflow recorded Standard at US$8.80 a month (660 credits) and Pro at US$32.56 (3,000 credits) after a discounted first month, with a limited-time 10% Kling 4.0 discount on Pro and above. Check the live plans page, because offers vary by region.
- No 4.0 API yet. Kling's developer platform showed "Kling 4.0 API Coming Soon", although its release notes say the MCP server and CLI already support 4.0.
- Commercial rights. According to Origami's review of Kling's terms, the free Basic plan does not permit commercial use, and subscription credits expire after 31 days.
- Language coverage. Kling's 4.0 language list names nine languages, including Chinese and English, followed by "and more". Cantonese voice and Traditional Chinese on-screen text are not explicitly named, so test them before a Hong Kong campaign.
One more honest note: rankings move fast. Origami reports that on LMArena's text-to-video board (to 21 September 2026), Kling's best placing was 28th, while it ranked higher for image-to-video. That pattern supports the keyframe approach: Kling tends to do better when you give it images to animate.
How can you test Kling 4.0 in 20 minutes?
Run one controlled test: take a product you already have photos of, build a three-beat, 12-second sequence with three keyframes, and generate it once with keyframes and once with a text-only prompt. Compare product accuracy, motion smoothness and stray text. The gap tells you whether keyframes are worth the prep time for your work.
Score both versions with this checklist:
- Did the product keep the same shape and logo at every beat?
- Did the motion between beats look natural, with no sudden jumps?
- Did any unrequested text, sign or watermark appear?
- How many credits did each version use?
- Would you show the result to a client without editing?
If the keyframe version wins on three of five, adopt the beat sheet as your team template. If it does not, you have learned which shots still belong with a camera. For still-image work on the same campaign, our Canva Craft layout-first guide pairs well with this method.
The real upgrade in Kling 4.0 is not the 30 seconds. It is that you can finally direct a clip instead of rolling dice on it. Plan the beats, pin the frames, and let the model handle the motion. We understand AI. We understand you better. With UD by your side, AI doesn't feel cold.
Reviewed by the UD AI team. Specifications and prices reflect Kling's user guide and plans page as reported by Wireflow (18 to 19 October 2026) and Origami (5 October 2026). Kling has not published per-video credit costs for 4.0, and features may change, so confirm in the Kling app before relying on them.
Can you beat AI at your own job?
Directing AI video is one skill. Knowing which parts of your work AI already does faster is another. Try UD's free AI Battle Staff challenge to find out, and when you are ready to turn techniques like this into a workflow that runs reliably every time, we'll walk you through every step, from tool setup to workflow design and deployment.