Kling Prompts for 4.0 and 3.0: Structure, Syntax and Examples to Copy

Which Kling version is current

Kling is Kuaishou's video generator, used through kling.ai on the web and in the app. Checked in November 2026: the current version is Kling 4.0, whose official guide came out at the end of October, while Kling 3.0 and 3.0 Omni, released in February 2026, are still available. The two generations are prompted differently, so start by deciding which one you're using.

• Kling 4.0: clips from 3 to 30 seconds (10 by default), in 720p, 1080p or 4K, at 21:9, 16:9, 1:1, 9:16 or automatic. Up to 15 reference assets and very long prompts.
• Kling 3.0 and 3.0 Omni: clips from 3 to 15 seconds, in 720p or 1080p, with multi-shot of up to 6 shots.
• Both generate audio with dialogue. Kling 4.0 does it in stereo with lip sync in nine languages; Kling 3.0 supports five.

These numbers change with every update. Before planning a project, check in the tool itself which lengths, resolutions and modes your account has.

How a Kling 4.0 prompt is structured

Kling's official 4.0 guide suggests a fixed order. Following it stops the model confusing what it should keep with what it should invent:

• Role of each asset: uploaded files are tagged @Image1, @Video1, @Element1… Say what each one is for and what must stay the same ('keep her face and coat').
• General settings: scene, visual style and camera, in concrete sentences.
• Timed sequence: what happens in each segment, written as '0–4 s: …', '4–10 s: …', until the full length is covered.
• Audio: dialogue in quotes, naming the language and describing the voice (age, tone, pace and emotion), plus ambient sound.

@Image1 is the main character's appearance: keep her face, her green apron and the pencil holding up her hair.
Scene: a small pottery workshop with shelves of drying cups. Warm interior light, photorealistic, mid-shot, the camera slowly pushes in.
0–4 s: she opens the kiln door and warm air shimmers.
4–10 s: she lifts out a cup glazed an unexpected deep blue and smiles.
Audio: the woman (in her 40s, warm voice, slow pace, pleasantly surprised) says in English: "Well, I didn't see that coming." Ambient: the kiln ticking as it cools.
No subtitles. No background music.

Kling 4.0 has two input modes that can't be combined: 'Omni Reference', the default, where you upload reference images, videos, voices or elements, and 'Keyframes', where you supply the first frame, the first and last, or several intermediate frames, but no reference assets. Pick one before you write, because the prompt changes.

Kling 3.0 and 3.0 Omni syntax

Kling 3.0 is still on offer and has its own conventions. For several shots in one clip (Multi-Shot or Custom Multi-Shot, up to six), write each shot with its length in brackets, or use timestamps such as '[00:00 - 00:02] Medium shot: …'.

Shot 1 (3s): wide shot, a courier in a yellow rain cape cycles down a rainy street at night, neon reflections on the wet pavement.
Shot 2 (4s): mid-shot, he brakes outside a bakery and checks a paper ticket, handheld feel.
Shot 3 (3s): close-up, he knocks on the glass door and grins.
Courier (cheerful tone, English): "Last delivery of the night!"

Note the dialogue format: the character first, then tone and language in brackets. In 3.0 Omni, when a character comes from a reference element, it is named with @ and written as '@Alan says, "…"'. Kling 3.0's modes are text-to-video, image-to-video, start and end frames, and 'Elements', which builds a character from two to four images or one video.

Dialogue, audio, references and negatives

• Dialogue: always in quotes, with the language named and the voice described. Kling 4.0 supports Chinese, English, Japanese, Korean, Spanish, Portuguese, German, French and Hindi; Kling 3.0 supports Chinese, English, Japanese, Korean and Spanish.
• References: call each file by its tag and give it a role. 'Use @Image1 as the character appearance and @Video1 as the camera movement reference' works better than 'use the photo'.
• Editing in 4.0: to change something in an uploaded video, write 'Edit @Video1 …, keep everything else unchanged' and name only what changes.
• Negatives: the Kling 4.0 guide advises against negative prompts because they can increase errors. It only recommends specific exclusions such as no subtitles or no background music. Describe everything else positively.
• On-screen text: give the exact wording, its size, its position and when it appears.

Edit @Video1: change the colour of the umbrella to bright yellow and make the rain heavier. Keep everything else unchanged: the man's face and clothes, the street, the camera movement and the timing.

Neither a negative prompt field nor 'motion brush' appears in the Kling 3.0 guides. If older tutorials mention them, check in the tool whether your version has them before relying on them.

Camera and movement in Kling

Kling understands plain-English film vocabulary. Ask for one move per shot or segment, and use clear verbs rather than adjectives:

• Moving in or out: 'slow push-in', 'the camera pulls back'.
• Going around: 'the camera circles the subject'.
• Handheld: 'handheld feel'.
• Angles: 'top-down', 'bird's-eye view'.
• Framing: 'wide shot', 'mid-shot', 'close-up'.

A street violinist plays on a rainy night, neon reflections on the wet pavement. Wide shot, then a slow push-in toward her hands on the strings. Cinematic, cool blue and pink tones. Ambient: rain on the pavement and the violin melody. No background music.

Common mistakes

• Vague wording like 'premium look' or 'epic' instead of describing light, framing and style.
• Uploading files without tagging them: if you don't say what @Image1 is and what to keep, the model decides for you.
• Mixing animation and photorealism in the same clip.
• Asking for on-screen text with no size, position or timing.
• Writing a long list of negatives for Kling 4.0.
• Carrying one version's syntax into the other, such as 'Shot 1 (3s)' inside a 4.0 prompt that already uses per-second segments.
• Asking for more length or resolution than your version allows: 3.0 tops out at 15 seconds and 1080p.

Scene: a bright, minimal kitchen counter with a ceramic teapot. Clean commercial look, soft daylight, static camera, close-up.
0–3 s: steam rises slowly from the spout.
3–6 s: the words "Tea time" appear in large white letters, bottom centre, and stay until the end.
Audio: ambient: a kettle clicking off and a quiet room.
No subtitles. No background music.

Using the PromptNow Kling prompt generator

If you'd rather not build the structure by hand, PromptNow's Kling prompt generator does it for you: pick the version (4.0, 3.0 or 3.0 Omni) and the mode, describe the idea, the subject and the files you'll upload, and you get a prompt for ChatGPT, Claude or Gemini that returns the final text in that version's syntax.

• It fits length, resolution and dialogue language to the chosen version's limits.
• It keeps the prompt apart from the settings you pick in the tool (length, aspect ratio, resolution, files).
• It explains what each part does and suggests two variants that each change just one thing.

Don't use real people, well-known voices, brands or copyrighted music, and label the video as AI-generated wherever the platform or the law requires it.