Kling 4.0 JSON prompt: a template you can copy

Why is everyone writing video prompts in JSON right now?

Because long clips fall apart when the prompt is one paragraph. A post on X this week about turning a simple idea into a JSON-style video prompt (scene, character, camera, timing, negatives) passed 23,000 views and 500 bookmarks in a day, and the replies are mostly people asking for the template.

The timing matters too. Kling 4.0 entered early access on September 30, 2026 with single takes up to 30 seconds and up to 10 keyframes. Kling 3.0, the version most accounts have today, tops out at 15 seconds. Thirty seconds is a lot of story to hold in one sentence. Breaking it into beats is the simplest fix.

One honest caveat. JSON is not magic syntax. Kling's announcement gives its formula in plain words: subject, action, setting, camera, lighting, pacing, sound. The JSON below uses exactly those slots, so nothing is lost if you later convert it back to prose.

What does a Kling 4.0 JSON prompt template look like?

Copy this and replace the placeholder text. Leave a key out rather than filling it with vague words.

{
  "subject": "who or what is on screen, with 2-3 fixed details",
  "setting": "place, time of day, weather",
  "beats": [
    { "time": "0-5s", "action": "opening state", "camera": "wide, static" },
    { "time": "5-12s", "action": "main movement", "camera": "slow dolly in" },
    { "time": "12-15s", "action": "ending pose or reveal", "camera": "close-up, hold" }
  ],
  "lighting": "direction, quality, color temperature",
  "style": "photoreal / 3D animation / film look",
  "pacing": "calm, steady, or fast cuts",
  "sound": "ambience, one effect, music or none",
  "avoid": "extra people, readable text, logo"
}

• subject: name two or three details that must not change, like a color or a dent. That is your identity lock.
• beats: one action and one camera move per beat. If a beat needs two camera moves, split it.
• pacing and sound: the two slots people skip, and the ones that make a long clip feel directed instead of random.
• avoid: keep it short. Three items work better than a long negative list.

Which example prompts can I copy and run?

Three prompts that cover the most common jobs. Each one runs on Kling 3.0 as written, since none goes past 20 seconds; trim the last beat if your plan caps you at 15.

{
  "subject": "a matte black insulated water bottle with a bamboo lid",
  "setting": "a granite boulder beside a mountain stream, early morning, light mist",
  "beats": [
    { "time": "0-4s", "action": "bottle stands on the rock, mist drifts past", "camera": "wide, low angle, static" },
    { "time": "4-10s", "action": "a hand lifts the bottle and unscrews the lid, steam rises", "camera": "medium shot, slow push in" },
    { "time": "10-15s", "action": "bottle set back down, water droplets catch the sun", "camera": "macro close-up, rack focus to the lid" }
  ],
  "lighting": "low golden sun from camera left, soft blue shadows",
  "style": "photoreal commercial, shallow depth of field",
  "pacing": "calm and deliberate",
  "sound": "running stream, a soft twist of the lid, no music",
  "avoid": "faces, brand names, text on the bottle"
}

{
  "subject": "a small round robot with one blue lens eye and dented orange paint",
  "setting": "a cluttered workshop at night, a single desk lamp",
  "beats": [
    { "time": "0-6s", "action": "the robot wakes up, lens flickers on, it looks around", "camera": "close-up, eye level" },
    { "time": "6-14s", "action": "it rolls across the desk toward a paper plane and nudges it", "camera": "tracking shot at desk height" },
    { "time": "14-20s", "action": "the plane slides off the edge, the robot leans over to watch it fall", "camera": "over the shoulder, then tilt down" }
  ],
  "lighting": "warm lamp pool, deep shadows, blue lens glow",
  "style": "stylized 3D animation, soft materials",
  "pacing": "playful, small pauses before each move",
  "sound": "servo whirs, paper rustle, a quiet beep at the end",
  "avoid": "humans, text, camera shake"
}

{
  "subject": "a ceramic cup of black coffee on a windowsill",
  "setting": "rainy afternoon, city rooftops blurred outside",
  "beats": [
    { "time": "0-5s", "action": "steam curls up, rain runs down the glass", "camera": "locked-off medium close-up" }
  ],
  "lighting": "soft gray window light, warm reflection on the cup",
  "style": "photoreal, muted palette",
  "pacing": "very slow, seamless loop",
  "sound": "rain on glass",
  "avoid": "hands, people, text"
}

Notice what stays the same across beats: the subject. Notice what changes: action and camera. That split is the whole trick.

How do you turn a photo into a Kling JSON prompt?

Image-to-video is where JSON pays off most, because the picture already fixes the subject and setting. Your prompt only has to describe motion.

1. Upload your still to the image to video prompt tool and choose Kling as the target.
2. Copy the JSON it returns. It includes shot, motion, camera, lighting and an audio cue for that frame.
3. In Kling, use the same image as the start frame and paste the JSON as the prompt. Keep the subject description short, since the image does that job.
4. If you have Kling 4.0 keyframes, add an end frame and write one beat for the move between them.

No photo yet? Make one with the Nano Banana prompt generator first, then animate it.

What should you do when Kling ignores part of the JSON?

Usually one of three things went wrong. The beats are too crowded, two camera moves compete in the same window, or the style key fights the image you uploaded.

Change one key per run. If the camera is the problem, edit only the camera values and regenerate. If it still drifts, flatten the JSON into sentences in time order, for example: "0 to 5 seconds: wide static shot of the bottle on a granite rock as mist drifts past. 5 to 10 seconds: a hand lifts it and the camera pushes in slowly." The prompt rewriter can do that conversion for you and keep every detail.

Questions people ask