Every “AI video generator pricing” list gives you sticker prices: $19 a month here, 1,000 credits there, $0.085 a second somewhere else. None of those numbers can be compared as written, which is the point of writing them that way.
There is one honest axis: dollars per second of finished output. We have put about $180 of metered production spend through these platforms and APIs; here is everything normalised onto that axis.
The table
| Route | Type | Effective cost | Basis |
|---|---|---|---|
| Seedance 2.0 Mini (BytePlus) | API | ~$0.03/s | Measured |
| MiniMax H3 (aggregator) | API | ~$0.08/s | Measured |
| Kling 3.0 (aggregator) | API | ~$0.08/s | Measured reservation |
| LumiYing — Kling 3.0 | Credit wallet | ~$0.08–0.10/s | Their credit table |
| Atlas Cloud — Seedance | API | ~$0.13/s flat | Advertised |
| Stivio (any model) | Credit wallet | ~$0.19/s | Their credit table |
| Higgsfield | Subscription | ~$0.09–0.19/s | Full utilisation of plan credits |
| Seedance 2.5, performance config | API | ~$0.30/s | Measured |
| LumiYing — Seedance 2.5 | Credit wallet | ~$0.29–0.37/s | Their credit table |
| fal.ai — Seedance | API | ~$0.47/s | Measured config |
| Dreamina / CapCut | Subscription | ~$0.13–0.16/s | Advertised tier |
| Runway (Seedance-class job) | Subscription | ~$0.36/s | Advertised tier |
Prices observed June–September 2026; “measured” means our own metered invoices, the rest are computed from the provider’s public pricing at full utilisation. Everything repricies monthly — treat the ordering as more durable than the digits.
Why the same model costs 10x more through one door
Notice that Seedance appears five times at five different prices. That is the whole story of this market: the same model is sold through at least seven doors, and the door decides the price. The pattern generalises:
APIs are the floor. $0.03–0.30 a second, but “some assembly required” is doing heavy lifting: console activation, asset hosting, retry logic. Cheap seconds plus your evenings.
Credit wallets are the middle. One balance, several models, browser interface. LumiYing lands near API parity if you buy its big pack; Stivio flattens everything to one rate — a discount on expensive models, a markup on cheap ones.
Subscriptions are the ceiling — with an asterisk. At full utilisation Higgsfield’s plans work out cheaper per second than most people expect. The catch is nobody achieves full utilisation: buy a big month, use half, and your real per-second price doubles. Subscription maths only works if your volume is boringly predictable.
The number the table can’t show you
Every figure above prices a generated second. You are buying usable seconds, and they are not the same number: our last full production batch was estimated at $81 from the rate card and billed at $115 once retakes and growing shot counts did their thing — a 42% overshoot on a batch we were actively supervising. Whatever route you pick, multiply by your own reject rate before believing a budget — the calculator does exactly that arithmetic.
One relief in all this: metered pricing is flat per second, with no per-clip surcharge — two four-second generations cost the same as one eight-second one. So budget by total seconds of output, and split shots wherever the edit wants them split; the meter does not care.
What to actually do
If you will write code, take the API floor: Seedance 2.0 Mini for volume, the model our bakeoff says fits your shot type for the rest. If you want a browser, pick by volume: occasional one-off videos — Stivio’s flat rate and free first video; regular multi-model work — LumiYing’s big pack at near-API rates; a need for visual direction tooling — Higgsfield, whose storyboard-and-camera-preset layer is the one thing here we have not been able to replicate in code, sized honestly against your real monthly output, not your ambitious one.
Check LumiYing’s credit packs →
Check current Higgsfield plans →
Disclosure: both links are affiliate links — a purchase through them earns AppMole a commission at no cost to you. The table above did not move for it; half the rows undercut both of these products.
