Kling 3.0 prompts work best when they read like a short scene direction, not a list of visual adjectives. Name the subject and setting first, then the camera move, then the action in order, then any dialogue or sound. Keep each generation to one scene of 3 to 15 seconds, and pick the model by the length, budget and audio you need. This guide was checked on 29 October 2026 against the fal and Morphic Kling 3.0 guides and Kenerate's own Kling v3 settings.
What does Kling 3.0 respond to in a prompt?
Kling 3.0 is built to read cinematic intent. fal's Kling 3.0 prompting guide says the model performs best when prompts are written as directions to a scene rather than a list of objects, with clear structure, explicit motion and deliberate shot language.
The Morphic Kling 3.0 guide lists the specs a prompt should plan around: durations from 3 to 15 seconds per generation, aspect ratios of 16:9, 9:16 and 1:1, native audio, and multi-shot output with up to six cuts in one video. Those four things decide what your prompt has to cover: the shot, the timing, the frame and the sound.
What structure should a Kling 3.0 prompt follow?
Use the same order every time. It makes prompts easier to edit, and it keeps the model from guessing which detail matters most.
| Part | What to write | Example |
|---|---|---|
| 1. Scene | Place and light, in one line | A rain-soaked street at night, neon reflecting on wet asphalt |
| 2. Subject | Who or what, described once and early | A woman in a yellow raincoat, dark hair tied back |
| 3. Action | What happens, in time order | She checks her phone, then looks up at the bus |
| 4. Camera | One named move | Medium shot, slow push-in as she looks up |
| 5. Sound | Ambience, effects, or a line of dialogue | Rain on the shelter roof, a bus brakes with a low hiss |
Here is that structure as one prompt:
A rain-soaked street at night, neon signs reflecting on wet asphalt. A woman in a yellow raincoat stands under a bus shelter, checking her phone. The camera holds a medium shot, then slowly pushes in as she looks up. Rain taps on the shelter roof. A bus pulls up with a low hiss.
This is our example, not a generated output. Swap any line and the rest of the prompt still stands on its own.
How should you describe camera movement and motion?
Name the camera's behaviour over time. fal's guide recommends describing how the camera moves across the clip, such as tracking, following, freezing or panning, rather than a general word like "cinematic". The same guide notes that long takes work well when the camera's relationship to the subject is clear, for example staying in a medium shot and freezing when the subject pauses.
Good motion prompts have three traits:
- One camera move per beat. "Slow push-in, then hold" is clearer than "dynamic camera movement".
- A stated relationship to the subject. "Follows her from behind" tells the model where the frame sits.
- A clear end state. "Settles on her face as the bus arrives" gives the clip a destination.
Avoid stacking several camera moves into a 5-second clip. The result is usually a muddy, jittery shot.
How do you write dialogue and sound?
With native audio on, the prompt has to say who is speaking and when. fal's guide gives four rules that hold across the examples:
- Label each speaker uniquely. Use the same label every time, such as
[Character A: Lead Detective]. Avoid switching between the name, a pronoun and a synonym. - Describe the action before the line. "The detective slides a folder across the table. [Lead Detective, calm but threatening]: 'Then explain why your fingerprints are here.'" tells the model who is doing what.
- Add a tone label for each voice. A raspy, deep voice and a clear, fearful voice are easier to tell apart than two generic "man" and "woman" labels.
- Use linking words for timing. "Immediately," or "Pause." marks a switch between speakers. Without it, lines can merge.
Sound effects work best when they are written as plain events in the scene: "paper scraping", "glass shattering", "a chair scrapes back". Morphic's guide says Kling 3.0 generates lip-synced dialogue in English, Chinese, Japanese, Korean and Spanish, with some accent and dialect control, so a language note in the prompt helps when the line is not English.
How do you plan a multi-shot clip?
Kling 3.0 can cut between up to six shots in one generation, according to both guides. Plan the cuts in the prompt rather than hoping the model will pick them:
Shot 1 (4 s): Wide shot of the platform, rain falling. The woman enters from the left. Shot 2 (3 s): Close-up on her phone, screen glowing. Her thumb hovers. Shot 3 (5 s): Medium shot, she looks up as the bus lights appear.
Label each shot, give its framing, subject and motion, and keep the character description identical in every shot. fal says the model keeps characters consistent when they are defined early and described the same way across shots. Morphic describes an auto mode, where the model plans the cuts, and a custom mode, where you set each shot's framing and duration. Check what the Kenerate studio exposes for your model before you write a storyboard, because the controls can differ by host.
How should image-to-video prompts differ?
In image-to-video, the input image is the anchor. fal recommends treating it as a locked first frame and writing the prompt around what changes: the camera move, a subtle movement, or an environmental shift. Do not describe the subject again in full, because the image already defines it. fal also notes that Kling 3.0 keeps text and signage from the source image, which makes it useful for product and branded shots.
In Kenerate, an image you add to a video run sets the start frame. Pick an aspect ratio that matches the frame, or the crop will change the composition.
Which Kling v3 model and length should you use?
Kenerate's Kling v3 entries share a 3 to 15 second range and the same three aspect ratios. They differ in price and in whether audio is charged.
| Model | Length | Aspect ratios | Credits per second, silent | Credits per second, with sound |
|---|---|---|---|---|
| Kling v3 Pro | 3 to 15 s | 16:9, 9:16, 1:1 | 33.6 | 50.4 |
| Kling v3 Standard | 3 to 15 s | 16:9, 9:16, 1:1 | 25.2 | 37.8 |
| Kling v3 4K | 3 to 15 s | 16:9, 9:16, 1:1 | 126 (sound included) | 126 |
Source: Kenerate's video model settings in src/lib/video-models.ts, as of 29 October 2026.
For a 5-second clip, that works out to:
- Pro, silent: 168 credits. Pro, with sound: 252 credits.
- Standard, silent: 126 credits. Standard, with sound: 189 credits.
- 4K: 630 credits, sound included.
The Starter plan includes 1,300 credits for $15 on the pricing page, so one 5-second Pro clip with sound uses about a fifth of it. A practical routine is to draft the prompt on Standard with the sound off, adjust the wording until the shot works, then run the keeper on Pro with sound or on 4K.
What are the most common Kling 3.0 prompt mistakes?
- Vague camera words. "Cinematic" does not tell the model where the camera goes. Name the move.
- Unlabelled speakers. Two lines of dialogue with no speaker labels often merge into one voice.
- Changing the character between shots. A new hair colour or outfit in shot three reads as a different person.
- Too many actions in one beat. Split a busy action into two short beats with a timing word between them.
- Asking for a 15-second clip with a thin idea. Long durations need a progression: what changes, how the camera reacts and how the scene ends.
Where should you go next?
If you are choosing between video models, read our Kling AI video generator page for the Kenerate Kling options, or the AI video generator page for the full model list. For a free-tier comparison, see is Kling 3.0 free. For camera language that carries across models, read AI video camera movement prompts. If you are writing for Seedance instead, our Seedance 2.5 prompt guide uses the same scene-first structure with different timing rules.
