A useful Kling 4.0 prompt tells the model what changes over time and what must stay consistent. Start with one subject, one legible action, and one camera plan. Add references when identity or product geometry matters, then judge the complete video rather than its best frame.
The examples below use Kling 4.0 on PoYo. The baker example is an existing completed PoYo task; the additional prompt templates are starting points to adapt, not guarantees of a particular result.
Describe the scene as shot direction
Use this order to keep instructions readable:
Subject and identity → action sequence → setting → camera
→ light and material → sound → continuity requirements
For example, replace “a cinematic chef doing something amazing” with a concrete sequence: the chef lifts a lid, steam escapes, and the chef sets it beside the pan. Specify whether the camera stays locked or follows the movement. Three competing camera moves in a short shot can obscure the action you actually want to test.
For identity, repeat the distinguishing details: a navy shirt, beige apron, short dark hair. For continuity, name the object that persists: “the same piece of dough” rather than a newly introduced loaf.
A real two-shot example
The following PoYo sample used multi_prompt, a total of five seconds, 720p, and 16:9. Recorded task: WXSD8J3D2MS5HI3U; recorded charge: 142.5 credits. It is a saved example, not a newly generated result each time this page loads.
{
"model": "kling-4.0",
"input": {
"duration": 5,
"resolution": "720p",
"aspect_ratio": "16:9",
"multi_prompt": [
{
"duration": 2,
"prompt": "Close-up at a lantern-lit night-market bakery. An adult male baker with short dark hair, a navy shirt and a plain beige canvas apron kneads one small piece of dough on a wooden counter. A soft puff of flour rises and settles with realistic weight and texture. Warm practical lighting, cinematic live action, natural hands and believable contact. No text, logos, subtitles or extra hands."
},
{
"duration": 3,
"prompt": "Cut to a medium view of the same adult male baker, identical short dark hair, navy shirt and beige canvas apron, at the same lantern-lit wooden bakery counter. He rounds that same dough gently with both hands, then places it on the wooden tray immediately beside him. Keep continuous action and consistent props across the cut. Subtle camera push-in, natural pacing and realistic anatomy. No text, logos or subtitles."
}
]
}
}
Watch whether the dough, hands and counter remain coherent across the cut. A correct request and a recorded task charge demonstrate execution; they do not establish a model-wide success rate or perfect physical accuracy.
Three camera templates
Locked camera for object interaction. Useful when you need to inspect hands, contact, and geometry:
One continuous five-second medium shot. An adult ceramicist lifts one
small cream bowl from a wooden table, pauses, and returns it to the same
spot. Keep the full bowl and both hands visible. Locked eye-level camera,
soft side light, realistic weight and shadows. Quiet studio ambience.
Tracking for locomotion. Describe the direction, pace, and frame boundary:
One continuous five-second low tracking shot beside a single red fox.
It takes two unhurried steps toward the right, pauses, and raises its
head. Keep all paws visible, with believable weight transfer and snow
compression. Match the reference animal's markings and proportions.
A restrained orbit for a product reveal. Separate the camera's movement from the product's movement:
The referenced bottle remains upright and stationary on a limestone
plinth. The camera makes one slow shallow arc from front-left toward
front-center. Preserve the cap, silhouette and color. Soft morning
light moves across the matte surface. Natural coastal ambience.
Dialogue and sound
Name the speaker, give the exact line, and state the tone. For a dialogue test, begin with one adult speaker and one short sentence. Then check lip timing, pronunciation, speaker identity, and whether extra words appeared. Adding several languages or overlapping speakers makes a failed result harder to diagnose.
PoYo's Kling 4.0 input schema does not expose a generate_audio switch or an audio-reference URL field. Express the intended soundtrack in the prompt and listen to the output. If music or subtitles appear despite your request, treat their removal as a review issue; a negative instruction is not an enforcement mechanism.
Troubleshoot one variable at a time
| Symptom | Next controlled change |
|---|---|
| The action happens too quickly | Reduce the action count or allow more time |
| The product changes shape | Add a clear product reference and simplify the camera |
| A character changes after a cut | Reuse explicit identity details or a subject reference |
| A hand loses contact with an object | Use a wider locked shot and slower interaction |
| Dialogue is hard to assess | Use one speaker and a shorter exact line |
Save the complete request with each result. PoYo currently accepts normal prompts up to 26,000 characters, but individual multi_prompt shot prompts are limited to 512 characters. Use either prompt or multi_prompt, never both. Per-shot durations must sum to 3–30 seconds and match an explicit top-level duration.
See the API guide for submission code, the reference guide for subject control, and the product-ad tutorial for a complete shot plan. The official Kling 4.0 user guide provides creator-app guidance; verify each feature against your API provider before copying its workflow.