Answer: a thirty-second AI video costs between about $0.30 and $6.30 depending on model and resolution — $0.60 for MiniMax H3 at 480p across two 15-second jobs, $1.80 for Kling 4.0 at 720p in one job, $3.60 for Seedance 2.5 at 720p, and $6.30 for H3 at 1080p in Callirra. Only some models render thirty seconds in a single take.
The duration ceiling is the part a price list hides. A per-second rate looks comparable until one model stops at ten seconds and the other runs to thirty, at which point you are comparing a shot to a sequence.
The table
| Model | 480p | 720p | 1080p | Single 30 s take? |
|---|---|---|---|---|
| MiniMax H3 | $0.30 | $0.60 | $0.90 | no (15 s max) |
| Seedance 2.5 | $1.50 | $3.60 | $6.30 | yes |
| Kling 4.0 | — | $1.80 | $2.67 | yes |
| Seedance 2.0 | $3.00 | $6.30 | $15.60 | no |
| Kling 4.0 video-to-video | — | $1.06 | $1.61 | yes |
Prices are settlement values at one credit = $0.01, for thirty seconds of output with audio where the model generates it. Higher plans lower every row proportionally, because they lower the price of a credit.
The two cheapest ways to get thirty seconds
MiniMax H3 at 480p — $0.30, in two takes. H3 caps at fifteen seconds, so thirty seconds is two renders joined in the edit: 15 s × 1 credit × 2 = 30 credits. The catch is the join, and 480p is not a delivery resolution.
Kling 4.0 at 720p — $1.80 in one job. Six credits a second for thirty seconds, with sound generated in the same render. It is six times H3's price, and it is one continuous take at a resolution people actually publish.
That comparison is the whole market in miniature: the cheapest per-second rate and the cheapest usable thirty seconds are different answers.
What the ceiling means in practice
| If the model stops at | You produce | You pay |
|---|---|---|
| 15 s | two takes plus a join | the same rate, twice |
| 10 s | three takes, two joins | the same rate, three times |
| 30 s | one take, no join | the rate, once |
The join is not free: it is an edit decision, a continuity risk, and another file to keep track of. Models with a 30-second ceiling are usually worth a small per-second premium when the shot genuinely needs to be continuous.
Resolution is the bigger lever than model choice
Compare the extremes: H3 at 480p is $0.30 for thirty seconds; Seedance 2.0 at 1080p is $15.60 for the same duration. That is a 52× spread, and most of it is resolution, not model.
Which suggests the ordering that actually saves money:
- Decide the delivery resolution first — 480p, 720p, 1080p.
- Among the models that offer it, pick the one with the duration ceiling you need.
- Only then compare per-second rates.
Doing it in the other order is how people end up with a cheap 480p plan they cannot deliver.
Where this table goes wrong
- Reference inputs. Video references add an input-duration charge on some routes; the table is output-only.
- Failed renders. Those release their reservation, so the table is a cost of finished work, not of attempts.
- Plan pricing. Settlement value is not your rate. Yearly top-plan credits are roughly a third cheaper per credit.
- Quality. Cheapest is not best; a $0.30 clip you cannot use costs more than a $1.80 clip you can.
FAQ
What is the absolute cheapest thirty seconds? MiniMax H3 at 480p: 30 credits, $0.30, as two fifteen-second jobs.
What is the cheapest single 30-second take? Kling 4.0 at 720p: 180 credits, $1.80.
Do these prices include audio? On Kling 4.0, H3 and Seedance 2.5, yes — audio is generated with the clip.
Why is 1080p so much more? Because it is priced per second at a higher rate, and thirty seconds multiplies that rate thirty times.
The per-second tables behind every row are on Callirra's credit-cost page, with model-specific detail on the Kling 4.0 page. If you find a cheaper thirty seconds that is also deliverable, that is the number worth publishing.
Top comments (4)
The duration ceiling is the part a price list hides — comparing a 10s model to a 30s model per second is comparing a shot to a sequence.
H3 needing two 15s takes vs Kling 4.0 doing 30s in one job is the clearest example of why the join matters as much as the rate.
The table is a great comparison of finished, accepted output — the caveat I'd add is retries. A "cheap" two-take H3 30 seconds can flip past Kling's single 720p take once you regenerate one of the 15-second halves or throw away a join that doesn't cut cleanly. So the usable cost is closer to: delivery resolution, then duration ceiling, then expected attempts, and only then the per-second rate.
The table finally puts the 30-second question in one place; the H3 480p vs Kling 720p trade is the one most people actually face.