Answer: a thirty-second Kling 4.0 take costs 180 credits at 720p ($1.80) and 267 at 1080p ($2.67), billed per output second. The safe way to get there is to buy the take in stages: a five-second test for the blocking, a fifteen-second render for the middle, and only then the full thirty — because every mistake in a long take is paid for thirty times.
Long takes are where budgets die, and not because the price is unfair. A six-second clip that is wrong costs 36 credits. A thirty-second clip that is wrong costs 180, and it takes six times as long to review.
The staged plan
| Stage | Duration | 720p credits | What you are deciding |
|---|---|---|---|
| 1 | 5 s | 30 | does the action read at all |
| 2 | 15 s | 90 | does the beat hold across the middle |
| 3 | 30 s | 180 | the take |
Total for one shot taken all the way: 300 credits, about three dollars. That is less than two failed thirty-second attempts, which is the normal cost of skipping stage 1.
What to write before you render anything
A thirty-second single take needs a beat sheet, not a description. Four lines:
0–6 s the courier reaches the bridge; camera tracks behind
6–14 s the ray descends alongside; flags lift
14–22 s the leap, then the dive between cliffs
22–30 s the waterfall clears; the wings catch and climbThen write the prompt as one paragraph that follows those lines in order, with the camera and the sound attached to the same beats. Kling 4.0 accepts up to 8,000 characters, which is more than enough for four beats — the limit is not what stops you.
Why the middle is the expensive part
In a five-second clip, the model can hold one event. In thirty seconds it must hold four, and the failure mode is not a broken frame — it is drift: the subject slows down, the camera forgets its direction, the soundscape stops matching the action.
That is why stage 2 exists at fifteen seconds. It is long enough to expose drift and short enough that finding it costs 90 credits instead of 180.
The cost of iteration, in full
| Iteration habit | Cost per shot |
|---|---|
| Render 30 s first, fix later | 180–540 credits |
| Test at 5 s, then 15, then 30 | 300 credits, one clean take |
| Test at 5 s, ship the 5 s | 30–60 credits |
The third row is worth saying out loud: many shots do not need thirty seconds. The ceiling is a capability, not a target.
1080p, and when to pay for it
1080p costs 8.9 credits a second — 48% more than 720p. Do the last stage at 1080p only if the deliverable is 1080p; rendering the test stages at 1080p is 48% extra for information you cannot see at that scale.
| Plan | Credits |
|---|---|
| 5 s + 15 s at 720p, 30 s at 1080p | 30 + 90 + 267 = 387 |
| All three stages at 720p, one 1080p re-render of the winner | 300 + 267 = 567 (only if needed) |
The first plan is almost always the better one: you decide at 720p, and you pay for 1080p once.
Where this workflow fails
- The shot needs a specific final frame. Use a keyframe pair instead of a bare text-to-video job; ordering
@first_frameand@last_frameremoves the drift at the end. - The sound must be exact. A thirty-second soundscape is one render; if the mix is wrong you re-render the whole thing, so write the audio beats as carefully as the action.
- You are chasing a reference. Do the reference exploration at five seconds on a cheaper route, then rebuild the shot at thirty.
FAQ
Can I extend a 5-second take instead of re-rendering? Not by asking the same job to continue — each render is a new job. Chain the takes in the edit, or render the full length once the blocking is right.
Is 30 seconds always the best length? No. It is the maximum, and the maximum is rarely the correct cut.
Do failed renders cost credits? Failed jobs release their reservation; you pay for files you can download.
Does the stage-1 test predict the 30-second take? It predicts the action, not the drift. That is why stage 2 exists.
The duration steps and the per-second price are on Callirra's Kling 4.0 page. If you have a thirty-second take that landed on the first try, the beat sheet is the interesting artefact.