# How much does an AI video API cost? Pricing per second in 2026
> What AI video APIs cost per second, what raises the price (resolution, audio, length, tier) and what one clip costs, with live Runway Dev prices.
Author: Runway · Published: 2026-10-26 · Updated: 2026-11-04 · Source: https://dev.runway.com/learn/ai-video-api-pricing
## Summary
Most AI video APIs charge between a few cents and about a dollar per second of generated video, at a rate set by the model, the output resolution and audio. A clip costs that rate times its length.\
\
This article covers billing units, what raises the price, what one clip or a thousand cost and how to spend less.
What an AI video API costs becomes a question once video is a feature you have to budget, often when you plan a launch or compare providers for one. Five things decide the answer: how the API bills, what it charges per second, what raises that rate, how long your clips run and how many you generate.
## 1. How do AI video APIs charge?
AI video APIs charge in five ways, and each one converts to a price per second of video, the figure to compare across providers:
1. **Per second of video**: a rate for each second of output, set by the model and its settings. A clip costs its rate times its length, so a 10-second clip costs twice a 5-second one.
2. **Per clip**: a fixed price for a video of a set length and resolution, such as 6 or 10 seconds. A clip you need at 8 seconds costs the 10-second price.
3. **Per token**: a price per million tokens, with tokens counted from the clip's resolution, frame rate and length.
4. **Prepaid units**: units bought in packages before use, often with an expiry date, then spent at a set number of units per second of video.
5. **Per minute or second of session**: real-time video, such as an interactive avatar streamed into a call, bills for how long the session lasts.
Runway Dev bills video per second of output, in credits drawn from one balance, at 1 credit = $0.01, and bills real-time avatar sessions with Runway Characters by their length. A per-second rate is the clearest of the five for a product team: a request's cost follows from the model, its setting and the clip's length, with no token count to estimate and no clip length to round up to.
## 2. How much does an AI video API cost per second?
AI video APIs typically cost from a few cents to about a dollar for each second of video they generate, at a rate that depends on the model and its settings. The table compares the best video models from five labs, each at its most useful resolution, on Runway Dev, on fal and Higgsfield, two other platforms that serve many labs' models, and on each lab's own API:
| Model | Runway Dev | fal | Higgsfield | Maker's API |
| -------------------------------------------------- | ------------------------------------------ | ------------------ | ------------------ | ------------------ |
| Seedance 2.5 ByteDance · 1080p | $0.68 | $1.164 | $1.137 | $0.569 |
| Kling 4.0 Kling · 1080p | $0.235 | $0.235 | $0.235 | $0.235 |
| Wan 3.0 Alibaba · 1080p | $0.20 | $0.20 | $0.20 | $0.20 |
| Gemini Omni Flash 1.1 Google · 720p | $0.10 | $0.10 | Not offered | $0.10 |
| Gen-4.5 Runway · 720p | $0.12 | Not offered | Not offered | Runway Dev |
Prices current as of November 4, 2026 on Runway Dev and November 4, 2026 elsewhere, at list price on every platform, before volume discounts, cashback or enterprise pricing. 1 credit = $0.01.
The fal and Higgsfield rates are the list prices on their model pages. The labs' rates come from their own pricing for [Seedance 2.5](https://docs.byteplus.com/en/docs/modelark/model-pricing), [Kling 4.0](https://kling.ai/dev/pricing), [Wan 3.0](https://www.alibabacloud.com/help/en/model-studio/wan3-0-video) and [Gemini Omni Flash 1.1](https://aistudio.google.com/docs/pricing). fal, Higgsfield and ByteDance bill Seedance 2.5 in tokens, so those figures are per-second estimates for a 16:9 clip.
No single video model is the best at every job, so the right model changes from one use case to the next, and a product that covers several use cases ends up calling several models. One platform that serves the best models keeps that to one key, one credit balance and one bill, and on these models Runway Dev costs the same as fal and Higgsfield or less. Every Runway Dev rate is on the [Runway Dev pricing page](https://dev.runway.com/pricing).
## 3. What makes an AI video API cost more?
Beyond its length, three things raise what a clip costs: the output resolution, native audio and the model tier.
### 3.1 Higher resolution raises the per-second rate
Resolution sets the rate on most video models, and each step up costs more per second. On Runway Dev, Seedance 2.5 costs 30 credits ($0.30) per second at 720p and 68 credits ($0.68) per second at 1080p, and Gemini Omni Flash 1.1 runs from 3.4 credits ($0.034) per second at 360p to 30 credits ($0.30) per second at 4K.
### 3.2 Native audio raises the rate on models where it is optional
Audio generated in the same pass as the video costs more than a silent clip on models that let you turn it off. Gemini Omni Flash 1.1 and Kling 4.0 generate sound with every clip, so audio is part of their per-second rate.
### 3.3 The tier sets the rate within a model family
Labs ship several tiers of one model at different rates. Wan 3.0 Prime generates faster than Wan 3.0 and costs more: on Runway Dev, 28 credits ($0.28) per second at 1080p, against 20 credits ($0.20) per second at 1080p for Wan 3.0.
## 4. How much do text-to-video, image-to-video and video-to-video APIs cost?
Text-to-video, image-to-video and video-to-video APIs all bill per second of output, and they differ in what they charge for the input you send.
### 4.1 Text-to-video API pricing is the output rate alone
A text-to-video API turns a prompt into a clip and bills its per-second rate times the clip's length, with nothing to pay for the prompt. On Runway Dev, text-to-video costs 68 credits ($0.68) per second at 1080p on Seedance 2.5, 23.5 credits ($0.235) per second at 1080p on Kling 4.0 and 12 credits ($0.12) per second of video on Gen-4.5.
### 4.2 Image-to-video API pricing usually matches text-to-video
An image-to-video API animates an image you send as the first frame, often with a last frame too, and most models bill it at their text-to-video rate. On Runway Dev, most models, including Seedance 2.5, Kling 4.0, Wan 3.0 and Gen-4.5, cost the same per second from an image as from a prompt, and a few, including Gemini Omni Flash 1.1, add a small charge for each input image.
### 4.3 Video-to-video API pricing depends on how each model bills the input
A video-to-video API edits, extends or restyles a clip you send, and models differ on whether that input adds to the bill. On Runway Dev, they fall into three groups:
- **Output only**: Seedance 2.0, Seedance 2.0 Fast and Seedance 2.0 Mini bill the clip they return, and a reference video adds nothing.
- **The length of the clip you edit**: an edit comes back as long as its source clip and bills at the model's usual rate, on Aleph 2.0, Runway's own video editing model, at $0.28 per second, and on Kling O3 Pro, Standard and 4K.
- **An added charge for the input**: Seedance 2.5 adds half its output rate for each second of reference video, MiniMax H3 bills reference video at its output rate, Gemini Omni Flash 1.1 adds a small charge per second of input video, and Kling 4.0 bills a lower rate on the output plus the longer of the input and the output.
## 5. How much does one AI video clip cost?
One clip costs its per-second rate times its length, plus any input charge. These are the costs of one 5-second and one 10-second clip from a text prompt on Runway Dev:
| Model and setting | 5-second clip | 10-second clip |
| ----------------------------- | --------------------------------------------- | ---------------------------------------------- |
| Gemini Omni Flash 1.1 at 720p | 50 credits ($0.50) | 100 credits ($1.00) |
| Wan 3.0 at 1080p | 100 credits ($1.00) | 200 credits ($2.00) |
| Gen-4.5 | 60 credits ($0.60) | 120 credits ($1.20) |
| Kling 4.0 at 1080p | 118 credits ($1.18) | 235 credits ($2.35) |
| Seedance 2.5 at 1080p | 340 credits ($3.40) | 680 credits ($6.80) |
A 10-second Seedance 2.5 clip at 1080p costs 680 credits ($6.80), and the same clip at 720p costs 300 credits ($3.00). A 10-second Kling 4.0 clip costs 235 credits ($2.35) at 1080p and 157 credits ($1.57) at 720p.
 on Runway Dev.')
```text
Locked-off wide shot of a golden wheat field at sunset under a dark storm front. A strong wind sweeps waves through the wheat while the heavy clouds roll slowly overhead. No people. Audio: wind gusting through the wheat, a low rumble of distant thunder.
```
## 6. How much does AI video cost at volume?
At volume, the bill is the price of one clip times the number of clips you generate, the takes you discard included, so a gap in the per-second rate becomes a gap in the monthly bill. These are the costs of 1,000 clips of 10 seconds for the best models from four labs, each at its most useful resolution, on Runway Dev, fal and Higgsfield:
| Model | Runway Dev | fal | Higgsfield | Savings vs fal | Savings vs Higgsfield |
| -------------------------------------------------- | --------------------------------------------------- | ------------------- | ------------------- | ------------------------------------------------------------- | ------------------------------------------------------ |
| Seedance 2.5 ByteDance · 1080p | $6,800 | $11,640 | $11,372 | $4,840 (42%) | $4,572 (40%) |
| Seedance 2.0 ByteDance · 1080p | $4,000 | $6,820 | $6,804 | $2,820 (41%) | $2,804 (41%) |
| Kling 4.0 Kling · 1080p | $2,350 | $2,350 | $2,350 | Same price | Same price |
| Wan 3.0 Alibaba · 1080p | $2,000 | $2,000 | $2,000 | Same price | Same price |
| Gemini Omni Flash 1.1 Google · 720p | $1,000 | $1,000 | Not offered | Same price | Not offered |
On the best models, Runway Dev costs the same as fal and Higgsfield or less. The widest gap is on Seedance 2.5 at 1080p, where 1,000 clips cost $4,840 (42%) less than on fal and $4,572 (40%) less than on Higgsfield. The fal and Higgsfield totals use the list rates on their model pages, checked November 4, 2026. Higgsfield bills Seedance 2.0 in tokens, so its figure is a per-second estimate for a 16:9 clip.
### 6.1 Retakes multiply the cost of a finished clip
A generation that finishes is billed whether you keep the clip or not, so a finished clip costs its price times the takes it took. At one kept take in three, a 10-second Gemini Omni Flash 1.1 clip at 720p costs 300 credits ($3.00) rather than 100 credits ($1.00). Budget for the takes you generate, which outnumber the clips you publish.
 in all, for the one take you keep.')
```text
Locked-off medium shot of a sunlit office desk with a wire letter tray at the right edge. A white paper airplane glides in from the left side of the frame, descends across the desk and lands in the tray. No people. Audio: a quiet office, the soft rustle of paper as it lands.
```
### 6.2 Failed generations can add to the bill
Failed generations add to the bill on providers that charge for them. On Runway Dev, a generation that fails because of an error on Runway's side is refunded automatically, so only the retry is billed.
## 7. What should you check before you budget?
1. The billing unit, converted to a price per second of output video.
2. The output resolution, since the rate rises with it.
3. Whether audio is included in the rate or billed on top of it.
4. The clip length your product needs.
5. The clips you need each month, and the takes behind each one you keep.
6. Charges for input video and reference images.
7. Whether a failed generation is charged.
8. Whether prepaid units or packages expire before you use them.
9. How many models one account, one key and one balance cover.
## 8. What does video generation cost on Runway Dev?
Runway Dev serves the best video models from the leading labs behind one API key: Seedance 2.5, Kling 4.0, Wan 3.0, MiniMax H3 and Gemini Omni Flash 1.1, alongside Runway's own Gen-4.5 and Aleph 2.0. Every model bills from one credit balance, at a fixed rate for each setting.
### 8.1 One key, one credit balance and one bill cover every model
Every model draws on the same credits, at 1 credit = $0.01. You buy credits for a project in the developer portal and spend them on any model at its rate, with no subscription. Runway app plans and credits are separate from Runway Dev credits.
### 8.2 Each model has a fixed rate, so a request's cost follows from its settings
Each video model has a fixed rate per second at each setting, so you can price a request before you send it. Every rate is on the [pricing page](https://dev.runway.com/pricing), and the dashboard and the usage API show the credits spent by model and by day.
### 8.3 A generation that fails on Runway's side is refunded
If an error on Runway's side fails a generation, its credits come back automatically, with nothing to claim.
### 8.4 The Enterprise plan adds volume pricing
The [Enterprise plan](https://dev.runway.com/enterprise) adds volume pricing and custom payment terms, set with Runway's sales team.
## 9. How do you cut AI video API costs on Runway Dev?
Runway Dev cuts video costs where they grow: the model each request pays for and the takes behind each clip you keep. A Model Router handles the first, and Recipes cut the second.
### 9.1 Pay only for the model each request needs, with fewer retakes
A [Model Router](https://dev.runway.com/model-routers) picks the model for each request from the priority you set, cost, quality or latency, and an optional credit ceiling per generation. It cuts cost two ways: it avoids over-provisioning, paying a top model's rate for a request a cheaper model handles as well, and a model matched to the request is less likely to need a retake. In [Runway's double-blind test of 250 image-to-video prompts](https://runway.com/news/developers/evaluating-runway-model-router), routing cut spend 29% at top-model quality when set to quality, and 66% at 95% of that quality with a $1 cost cap. A routed request bills at the rate of the model it picks, with no routing fee.
### 9.2 Get production-ready assets in one API call, no iteration required
A [Recipe](https://dev.runway.com/recipes) is a Runway-built endpoint for one production outcome, such as a product ad or a localized campaign, with the model selection and prompting that outcome takes packaged in. You send the inputs the Recipe lists, with no prompt to engineer and no models to chain, so fewer takes stand between a request and a clip you publish. Runway can update the models and prompting behind a Recipe while its endpoint stays the same.
## FAQ
**Do you pay for AI video generations you don't keep?**
Every generation that finishes is billed, whether you keep the clip or not, so retakes add to the bill. Runway Dev bills each finished generation at its model's rate and returns the credits of a generation that fails because of an error on its side.
**Does audio cost extra in an AI video API?**
Audio costs extra on models where it is optional, since generating sound in the same pass adds to the per-second rate. Models that always generate sound, such as Gemini Omni Flash 1.1 and Kling 4.0, include it in the rate.
**Does resolution change the price of an AI video API?**
Resolution sets the rate on most video models. On Runway Dev, Seedance 2.5 costs 30 credits ($0.30) per second at 720p and 68 credits ($0.68) per second at 1080p, and Gemini Omni Flash 1.1 runs from 3.4 credits ($0.034) per second at 360p to 30 credits ($0.30) per second at 4K.
**Do AI video APIs charge for failed generations?**
Refund policies differ by provider. Runway Dev returns the credits of a generation that fails because of an error on its side, automatically.
**What is a credit in an AI video API?**
A credit is a prepaid unit that a platform sells at a fixed price and quotes its rates in. On Runway Dev, 1 credit = $0.01, and one balance of credits pays for every model.
**How much does the Runway API cost?**
Runway Dev, Runway's API platform, charges per second of video from one credit balance. Gen-4.5 costs 12 credits ($0.12) per second of video, and the same balance pays for Kling, Seedance, Wan and Gemini Omni Flash at their own rates, listed at dev.runway.com/pricing.
**Can you know what an AI video request costs before you send it?**
Each video model on Runway Dev has a fixed rate per second at each setting, so a request's cost follows from the model, the setting and the clip's length. A 5-second Gen-4.5 clip costs 60 credits ($0.60).
**Do you need a subscription to use an AI video API?**
Runway Dev is pay as you go, with no subscription. You buy credits for a project and spend them on any model at its rate. Runway app plans and credits are separate from Runway Dev credits.
**Can you pay for an AI video API through AWS?**
Runway Dev is on AWS Marketplace, so its usage can go on your AWS bill. A subscription through the public listing starts once Runway approves it, and existing credits in the project are spent first.
**Is there volume pricing for AI video APIs?**
Runway Dev offers volume pricing and custom payment terms on its Enterprise plan. Talk to sales at dev.runway.com/enterprise.