Kling 4.0 changes how you should prompt. According to Kling's announcement (September 30, 2026), it generates up to 30 seconds in a single take, accepts up to 10 keyframe images, takes up to 15 reference assets (images, videos and subjects with voices), outputs two-channel stereo audio with improved lip sync, and understands prompts of up to 8,000 tokens. Early access started September 30, with a wider rollout announced for October 2026.
This guide turns those features into practical prompt patterns. If you are new to video prompting, start with our text to video prompt ideas. Note: Kling 4.0's API was not yet publicly available at the time of writing, so third-party platforms, including TextVidAI, may not offer it yet.
What changed for prompting in Kling 4.0
- Longer clips need a story, not a moment. At 30 seconds you should describe a beginning, a development and an end, or the model will pad the scene.
- Keyframes replace guesswork. With up to 10 keyframe images you can pin poses, expressions and transitions at specific beats.
- References carry identity. Up to 10 images, 5 videos (30 seconds total) and 7 subjects can be combined, so describe what each reference is for.
- Audio is part of the prompt. Kling 4.0 supports many languages and accents (including American, British and Indian English), so specify language, accent and tone for any dialogue.
- More text is allowed. 8,000 tokens means you can write a short shot list, but clarity still beats length.
The prompt structure that works
Write prompts in this order. It mirrors how a director briefs a crew:
- Format: duration, aspect ratio (Kling 4.0 also supports 21:9), single take or multi-shot.
- Subject: who or what, with fixed visual details.
- Action over time: what happens at the start, middle and end.
- Camera: movement and framing per beat.
- Audio: dialogue in quotes, sound effects, ambience, music.
- Style and light: look, lens, color, time of day.
- References: what each attached image, video or subject controls.
Template 1: 30-second single take
30-second single continuous take, 16:9. A street-food cook in a white apron prepares noodles at a night market stall. 0–10s: close-up of hands pulling dough, steam rising; camera slowly pushes in. 10–20s: she tosses noodles into a wok, flames flare; camera arcs left to reveal the busy market behind her. 20–30s: she hands a bowl to a customer and smiles; camera pulls back to a wide shot of the stall. Audio: sizzling wok, crowd chatter, distant music; she says in American English, "Careful, it's hot." Warm tungsten light, shallow depth of field, documentary style.
Template 2: keyframe-driven transformation
Use keyframe 1 as the opening frame, keyframe 2 at the midpoint and keyframe 3 as the final frame. A paper airplane folds itself from a flat sheet on a desk (keyframe 1), lifts off and flies through an open window (keyframe 2), and lands on a rooftop garden at sunset (keyframe 3). Smooth continuous camera following the airplane. Soft wind and paper rustle; light acoustic music.
Tip: keep lighting and color temperature consistent across keyframes, or the model has to invent a transition between them.
Template 3: product ad with references
15-second product ad, 9:16. Product: the bottle in reference image 1, keep the label and logo exactly. Setting: the bathroom in reference image 2. 0–5s: macro shot of water droplets on the bottle. 5–10s: a hand picks it up and sprays; mist catches the morning light. 10–15s: bottle placed on marble, text-free final frame for an end card. Audio: soft spray sound, calm piano.
For more product ideas, see how to make AI product ads that convert.
Template 4: dialogue scene
Two-shot, 12 seconds, 16:9. A father and teenage daughter sit in a parked car in the rain. Daughter, nervous, in British English: "I didn't get in." Father pauses, then smiles: "Then we'll try again next year." Camera slowly pushes in on the daughter. Audio: rain on the roof, windshield wipers, no music. Natural, muted colors.
Common mistakes and fixes
- Clip feels padded: add a clear event in each third of the clip.
- Character drifts: attach a subject reference and repeat two or three fixed traits (hair, clothing color) in the prompt.
- Wrong voice or accent: state language and accent next to each line of dialogue.
- Camera ignores you: give one camera instruction per beat, not a list of moves at once.
- Unwanted text in frame: say "no on-screen text" when you plan to add titles later.
For a deeper look at consistency, read how to keep characters consistent in AI video.
Kling 4.0 vs Kling 3.0 for prompting
Kling 3.0 tops out at around 15 seconds with multi-shot storyboards; Kling 4.0 doubles single-take length and adds multi-keyframe control and larger reference sets. If your platform only offers earlier Kling versions, the same structure still works: shorten the action to fit the clip length. Our Kling 2.6 explainer covers the earlier generation.
Try it
You can practise this prompt structure today with the Kling and Veo models available in the TextVidAI generator, then move to Kling 4.0 as it becomes available.
FAQ
How long can a Kling 4.0 video be?
Up to 30 seconds in a single generation, according to Kling. Kling also announced video extension up to 2 minutes as "coming soon."
How many keyframes does Kling 4.0 support?
Up to 10 keyframe images per generation.
Does Kling 4.0 generate sound?
Yes. It generates audio with two-channel stereo and improved lip sync, in many languages and accents.
Sources
Last updated: October 9, 2026.