Answer: Kling 4.0 exposes four controls that matter — duration (3–30 s in one-second steps), resolution (720p or 1080p), aspect ratio (16:9, 9:16, 1:1, 21:9) and references (an image, an ordered keyframe pair, or a video). Duration and resolution set the price; references and aspect set what the shot can be.
Everything else in the prompt is craft. These four are the ones that determine whether the job is even the shot you wanted.
1. Duration: the only control that scales the bill linearly
Kling 4.0 accepts 3 to 30 seconds in one-second steps. At 6 credits per second on 720p, the arithmetic is direct:
| Duration | 720p | 1080p |
|---|---|---|
| 3 s | 18 credits | 27 credits |
| 5 s | 30 | 45 |
| 10 s | 60 | 89 |
| 20 s | 120 | 178 |
| 30 s | 180 | 267 |
The design question is not "how long can I make it" but "does this beat need the length". A 30-second take is one continuous shot, which means the camera has nowhere to hide: if the blocking or the light is wrong, it is wrong for thirty seconds.
2. Resolution: 720p is a drafting tier, not a compromise
720p costs 6 credits a second, 1080p costs 8.9 — about 48% more. The practical split:
- 720p for exploration. A five-second test is 30 credits; twenty of them cost less than one 30-second 1080p take.
- 1080p for the take you keep. The difference is visible in edges and fine texture; it is not visible in blocking.
Neither tier reaches 4K. If a client spec says 4K, this is not the model — Seedance 2.0 carries that tier.
3. Aspect ratio: decide before the prompt, not after
Four options: 16:9, 9:16, 1:1 and 21:9. The ratio changes composition, not just framing — a 21:9 letterbox wants different headroom than a 9:16 vertical, and a prompt written for one will fill the other badly.
| Delivery | Ratio | Note |
|---|---|---|
| YouTube, film | 16:9 | the default |
| Reels, Shorts, TikTok | 9:16 | write for vertical headroom |
| Feed post | 1:1 | safest for mixed placements |
| Trailer, anamorphic look | 21:9 | needs deliberate negative space |
4. References: one image, two keyframes, or a clip
This is where Kling 4.0 is more flexible than most models at this price:
- Image to video. One still establishes the look; the model animates from it.
- Keyframes. Two ordered stills fix the start and the end of the move. The published pavilion example uses
@first_frameand@last_frameto hold architecture, light direction and horizon steady across a ten-second dolly. - Video to video. A clip becomes the reference for route and motion, and the prompt changes the light or the time of day — the blue-hour restyle example keeps the glide and swaps the ambience.
Rule of thumb: a keyframe pair is for a move with a known destination; a video reference is for a move you already shot and want to restage.
How the four interact
| If you change | What also has to change |
|---|---|
| 16:9 → 9:16 | the prompt's composition language, and the reference crop |
| 720p → 1080p | nothing in the prompt; the price rises ~48% |
| 5 s → 30 s | the beat sheet — a long take needs events across its length |
| one image → keyframe pair | the prompt becomes a description of the move between them |
Where people get it wrong
- Choosing 30 seconds to "get more". You pay 6× the 5-second price for one shot that must carry a scene.
- Feeding a 16:9 keyframe into a 9:16 job. The model crops or letterboxes; either way the composition you approved is gone.
- Writing a silent prompt and expecting sound. Kling generates audio; say what to hear.
- Assuming a video reference locks the output. It locks route and motion, not look — that is the point of restyling.
FAQ
What is the default? 720p, 5 seconds, 16:9.
Can I render longer than 30 seconds? Not in one job. Chain takes with the last frame of one as the first frame of the next.
Does a keyframe pair cost more? No. Price follows duration and resolution.
Does the reference have to come from my library? It can be an uploaded image or an owned asset; either way it stays private to your account.
The controls and their options are on Callirra's Kling 4.0 page; pick the pair of keyframes before you write the move, and the rest of the prompt gets shorter.