Kling Multi-Shot: How Kling 3.0 Storyboards Work
How Kling multi-shot works in Kling 3.0: the official shot syntax inside the prompt, six-shot and 15-second limits, and how to write a shot list that holds.
By OceanToken Team

Editor's note: Kling multi-shot turns a clip generator into a scene generator. We think it is the best reason to use Kling 3.0, and also the easiest feature to misuse. People write one long paragraph and hope for cuts. Kuaishou's API asks for something stricter: a numbered shot list with a duration on every shot, inside the prompt itself. Write it that way and the cuts land where you planned them.
Short answer: Kling multi-shot turns one prompt into a clip with cuts. On Kuaishou's Kling 3.0 API you write up to six shots inside prompt as shot n, m, words;, where m is seconds. Each shot needs at least 1 second and 512 characters or fewer, and the shot durations must add up to the clip length, 3 to 15 seconds.
What Kling multi-shot does
Without multi-shot, a Kling generation is one continuous take. With multi-shot, you describe a sequence as separate shots and the model returns a single video with cuts between them. Because the model sees the whole sequence in one request, it can keep a character, a prop and the lighting consistent across cuts in a way that separate calls cannot.
Not every Kling version does this. Kuaishou's video capability map , checked 2026-11-06, lists multi-shot for text-to-video and image-to-video on Kling 4.0, 4.0 Flash, 3.0 and 3.0 Turbo. Kling O1 does not support it, 2.6 and 2.5 Turbo are single-shot only, and 3.0 Omni offers it only under Omni Reference. So before you write a storyboard, check that the model you call is on that list.
The Kling 3.0 multi-shot format
On the Kling 3.0 text-to-video reference , there is no separate shot field. The shot list is plain text inside prompt, in the form shot n, m, words;. The rules:
| Rule | Value |
|---|---|
| Where shots go | Inside prompt, each shot ending with ; |
n, shot number |
1 to 6, so six shots at most |
m, shot length |
Whole seconds, at least 1 per shot |
| Shot durations | Must add up to the total video duration |
| Total duration | settings.duration, 3 to 15 seconds, default 5 |
words, shot prompt |
512 characters or fewer per shot |
| Whole prompt | 3,072 characters or fewer, 2,500 or fewer recommended |
| Switch | settings.multi_shot, default true; false means multi-shot prompts give single-shot output |
Two practical rules follow. Set settings.duration explicitly whenever you send shots, because the default is 5 seconds and three 4-second shots will not add up to it. And count before you submit: four 4-second shots is 16 seconds, over the 15-second cap.
If you call Kling through a third-party host rather than Kuaishou, do not assume this syntax carries over. A host may expose shots as structured fields with its own names and limits, so read that host's schema first.
How to write a Kling multi-shot prompt
Each shot needs five things. If one is missing, the model guesses.
| Element | Example | Why |
|---|---|---|
| Subject | "keeper in a yellow oilskin" | Identity anchor; repeat it every shot |
| Action | "climbs with one hand on the rail" | One verb per shot |
| Camera | "tracking from behind" | One move only |
| Duration | 4 s | Sets the pacing |
| Light and style | "warm lamp glow and film grain" | Keeps the cuts consistent |
Repeat the identity words. Do not write "the same man." Restate a short description in each shot: clothing, age and one prop.
Use one camera move per shot. "Dolly in while orbiting and tilting up" tends to blur. Pick one.
Vary the framing. Alternate size and angle between shots: wide, then close, then low. Two similar shots in a row look like a glitch, not a cut.
A three-shot example
Here is a request body for Kuaishou's Kling 3.0 endpoint, using only documented fields. We kept commas out of the shot text so the separators stay unambiguous; the docs do not say whether commas inside a shot cause trouble.
{
"prompt": "shot 1, 4, Wide shot of a lighthouse keeper in a yellow oilskin walking up a rocky path at dusk with a static camera and warm lamp glow and film grain; shot 2, 4, Close-up of the keeper in a yellow oilskin gripping the iron rail and climbing while the camera tracks from behind with warm lamp glow and film grain; shot 3, 4, Low angle on the keeper in a yellow oilskin lighting the lamp at the top with a slow push-in and warm lamp glow and film grain;",
"settings": {"duration": 12, "aspect_ratio": "16:9", "resolution": "1080p", "multi_shot": true}
}
Total: 12 seconds, three shots of 4 seconds, and settings.duration set to match. The subject and style words repeat in every shot. The create, poll and download steps are the same as for any Kling task, covered in our Kling API tutorial.
Where multi-shot breaks
- Joins. Cut points are the weakest part of the output. Ask for plain cuts rather than dissolves, and plan to trim.
- Drift. A prop in shot 1 may change color by shot 3. Name it in each shot.
- Dialogue. On Kling 3.0,
settings.audiodefaults tooff. If you switch it tonative, a line split across shots may sound uneven, so keep one short line per shot.
Native multi-shot or stitched clips?
You can still make each shot as its own call and join them in an editor.
| Native multi-shot | Separate calls, stitched | |
|---|---|---|
| Identity across shots | Model sees the whole sequence | Each call starts fresh |
| Control per shot | One request, shared settings | Full control per shot |
| One bad shot | Rerun the whole sequence | Rerun only that shot |
| Joins | Made by the model | You choose cut or fade |
Rerunning a full sequence costs more than rerunning one short shot. Kling bills video by length: its task results report an output_duration that the Kling 4.0 reference says is used for billing.
Our take: Use native Kling multi-shot for the first draft, because the model sees all shots at once and keeps identity better. When one shot keeps failing, stop rerunning the whole sequence. Regenerate that shot alone and cut it in.
How to test this on your workload
- Pick one three-shot storyboard from a real project.
- Run it five times as native multi-shot at 12 seconds.
- Run it five times as three 4-second calls and stitch them.
- Score identity drift, join quality and the reruns needed for one usable take.
- Note the total seconds billed for each approach.
Running it on OceanToken
OceanToken lists kwaivgi/kling-v3.0-pro and kwaivgi/kling-v3.0-std, plus kwaivgi/kling-v3.0-omni and kwaivgi/kling-video-o1. Use base URL https://api.oceantoken.ai/v1 with videos.create_and_poll and videos.download_content from any OpenAI-compatible SDK.
The /kling-api page does not document multi-shot yet. Since Kuaishou's format lives inside the prompt text, send one short test with the shot n, m, words; syntax in prompt and seconds matching the shot total, and check whether the clip comes back with cuts before you build on it. The reason to use OceanToken here is one key and one balance for Kling, other video models and text models; if you are still deciding between routes, see official Kling API vs a multi-model gateway. See the current rate on /kling-api.
FAQ
How many shots can Kling make in one video? Up to six on Kuaishou's Kling 3.0 API, numbered 1 to 6 in the prompt.
Can I set the length of each shot? Yes. Each shot has its own duration of at least 1 second, and the shots must add up to the total, which is at most 15 seconds on Kling 3.0.
Which Kling versions support multi-shot? Kuaishou lists Kling 4.0, 4.0 Flash, 3.0 and 3.0 Turbo for text-to-video and image-to-video.
Why does my multi-shot request come back as one take? Check that settings.multi_shot is not false, that the model supports multi-shot, and that the shot durations add up to settings.duration.
Sources
- Kling API docs: Kling 3.0 text to video (checked 2026-11-06)
- Kling API docs: video capability map (checked 2026-11-06)
- Kling API docs: Kling 4.0 text to video (checked 2026-11-06)
- OceanToken Kling API page (checked 2026-11-02)
Get your OceanToken API key
- Create an OceanToken account and an API key.
- Copy the snippet below, paste your key, and run it.
- Check the current rate on /kling-api.
curl https://api.oceantoken.ai/v1/videos \
-H "Authorization: Bearer $OCEANTOKEN_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "kwaivgi/kling-v3.0-pro", "prompt": "Wide shot, a lighthouse keeper in a yellow oilskin walks up a rocky path at dusk"}'
More from the blog
A Unified API for Chat, Image and Video
Building on a unified API for chat, image and video: sync vs async calls, billing units, retries, timeouts and storage, with three calls and a launch checklist.
Kimi K3 API International Access Guide
Use the Kimi K3 API internationally: sign up with email, pay by card, create one key and call moonshotai/kimi-k3 with an OpenAI SDK. Steps, code and checks.
Call Tencent Hy3 With the OpenAI SDK
Call Tencent Hy3 with the OpenAI SDK outside China: base URL, model id, streaming, tool calls, reasoning_effort values, Hy3 vs Hy3 preview, and token billing.