Published October 2026 · 10 min read
Ask what a 30-second AI video costs and you will get answers ranging from “about a dollar” to “I spent two hundred and got nothing usable.” Both are honest. They are describing different things.
The confusion comes from treating “30 seconds of video” as a single unit. It is not. It is a duration multiplied by three variables that most people forget to count: resolution, audio, and retries. Get those wrong and your budget estimate will be off by a factor of ten or more.
This article does the arithmetic in the open, using the same public API list prices that power our cost calculator. All figures are USD per second of generated output, compiled from public pricing pages as of July 2026. Confirm on the vendor's page before any large job — these move.
Here is the spread across ten models at 720p and 1080p. This is the number people quote when they say AI video is cheap, and for the low end they are right.
| Model | 720p / sec | 1080p / sec | Audio |
|---|---|---|---|
| Kling 3.0 (API resellers) | $0.04 | $0.08 | No |
| Wan 2.6 | $0.05 | $0.05 | No |
| Runway Gen-4 Turbo | $0.05 | $0.10 | No |
| Veo 3.1 Lite | $0.05 | $0.10 | Included |
| Veo 3.1 Fast | $0.10 | $0.20 | Included |
| Sora 2 | $0.10 | — | Included |
| Runway Gen-4.5 | $0.12 | $0.24 | No |
| Seedance 2.5 | $0.20 | $0.40 | Included |
| Sora 2 Pro | $0.30 | $0.50 | Included |
| Veo 3 (original) | $0.75 | — | Included |
The spread from cheapest to most expensive is roughly 19× at 720p. That single fact explains most of the contradictory numbers you see online. Someone quoting “$1.20 for 30 seconds” is using Kling at 720p. Someone quoting “$22.50” is using Veo 3. Neither is lying.
Moving from 720p to 1080p is not a linear upgrade. On several models it roughly doubles the per-second rate:
Wan 2.6 is the outlier and worth remembering: it generates natively at 1080p for the same rate. If your output genuinely needs 1080p and you are on a tight budget, that single line item can halve your spend.
A practical note: 1080p is not always necessary. Vertical short-form is viewed on a phone at roughly 1080×1920, so 1080p is the correct target there. But for a shot that will be cropped, reframed, or shown as a background element, 720p upscaled in the edit is usually indistinguishable and half the price.
Models marked “Included” generate synchronised audio — dialogue, ambient sound, effects — as part of the render. Models marked “No” produce silent video and you add audio separately.
This looks like a simple feature difference and it is actually a budget question. If you are producing talking-head content or anything where lip-sync matters, a silent model means you are paying twice: once for the video, once for a separate voice/lip-sync pass. On the cheap end that can double the effective cost of the shot.
Conversely, if your video is music-driven with no dialogue — product b-roll, aesthetic montage, most faceless channel content — paying for native audio is paying for something you will mute in the edit. In that case the silent models are strictly better value.
This is where the naive calculation falls apart. You do not pay for the video you use. You pay for every generation you attempt.
In practice, a shot that reads as final usually takes two to four attempts. The first establishes composition, the second fixes a detail, the third may be needed because the model drifted on the face or the hands. Professional workflows budget three generations per finished shot as a baseline — and that is on models you know well.
So the honest formula for a 30-second video is:
cost = 30 seconds × (retries per shot) × (price per second)
Which means a 30-second video is really 90 seconds of paid generation at three attempts per shot. Here is what that does to the numbers at 1080p:
| Model (1080p) | One pass (30s) | Realistic (90s) | Heavy retry (150s) |
|---|---|---|---|
| Wan 2.6 | $1.50 | $4.50 | $7.50 |
| Kling 3.0 | $2.40 | $7.20 | $12.00 |
| Runway Gen-4 Turbo | $3.00 | $9.00 | $15.00 |
| Veo 3.1 Lite | $3.00 | $9.00 | $15.00 |
| Veo 3.1 Fast | $6.00 | $18.00 | $30.00 |
| Runway Gen-4.5 | $7.20 | $21.60 | $36.00 |
| Seedance 2.5 | $12.00 | $36.00 | $60.00 |
| Sora 2 Pro | $15.00 | $45.00 | $75.00 |
“Heavy retry” is five attempts per shot, which happens when you are learning a new model or chasing a difficult composition. Note that at the top of the table, a single 30-second clip can cost more than a month of a consumer subscription. That comparison is the next trap.
Everything above is API pay-as-you-go pricing, which is the cleanest way to compare models because you pay for exactly what you generate. Consumer subscriptions bundle a credit allowance at an effective rate that is often much lower — and that is genuinely the better deal if you are a solo creator producing a handful of clips a week.
The subscription becomes the worse deal when you hit the ceiling: credits run out mid-project, the top-up rate is worse than the bundled rate, and you are locked to one model for everything. The break-even is roughly where your monthly generation volume exceeds what the bundled credits cover — below that, subscribe; above it, go API and route each shot to whichever model is best value for that shot.
If you are converting between credits and dollars to compare a subscription against API rates, the credits converter does that arithmetic directly.
Generation is the visible cost. Three others are real and routinely ignored:
Abstract ranges are easy to nod at and hard to act on. Here is a real allocation for a 30-second product video, split into six shots of five seconds each, with three attempts budgeted per shot — 90 seconds of paid generation in total.
| Shot | What it is | Model (1080p) | Paid seconds | Cost |
|---|---|---|---|---|
| 1 | Wide establishing shot, no faces | Wan 2.6 | 15 | $0.75 |
| 2 | Product close-up, static | Veo 3.1 Lite | 15 | $1.50 |
| 3 | Product detail, slow push in | Veo 3.1 Lite | 15 | $1.50 |
| 4 | Person using it, needs audio | Veo 3.1 Fast | 15 | $3.00 |
| 5 | Texture b-roll, silent | Kling 3.0 | 15 | $1.20 |
| 6 | Closing shot, silent | Kling 3.0 | 15 | $1.20 |
| Total | $9.15 | |||
Compare that with two single-model approaches for the same 90 paid seconds:
The routed version costs roughly half of the cheapest single-model option and a fifth of the most expensive. Nothing was sacrificed: the establishing shot genuinely did not need a premium model, and the one shot that needed native audio got the model that provides it. This is what “route by shot” means in practice, and it is the single highest-leverage change most people can make to their spend.
For a 30-second video at 1080p with realistic retries, the honest range is $4.50 at the low end to $45 at the high end, with most work landing between $9 and $22. That is the answer to the question, and it is deliberately a range rather than a number — because the three multipliers above are choices, not constants.
The comparison that matters is not AI model versus AI model. It is AI generation versus the alternative. A comparable 30-second spot from a freelance videographer, with a shoot day, talent and editing, typically runs several hundred dollars at minimum. Our AI vs freelancer cost breakdown puts real numbers on that gap — and it is the reason this whole category exists.
Two caveats worth repeating before you build a budget on any of this. First, these are list prices compiled in July 2026, and every vendor in the table has changed its rates at least once in the past year — treat the absolute numbers as a planning baseline rather than a quote. Second, regional pricing, sales tax and promotional credits are excluded from every figure here. If you are paying in a currency other than USD, the effective rate per second can differ noticeably from the published API price, and that difference is usually larger than the gap between two adjacent models in the table.
Run your own figures in the cost calculator before the next project. Enter the duration you actually need, at the resolution you will actually deliver, with the retry count you actually experience — not the one you hope for.