Cost reference
How much does AI video cost?
Every AI video tool publishes a monthly price. Almost none publish the number that decides your bill: what a single generation costs. That number is not a property of the model alone — it is model × settings, and it moves with resolution, duration and whether you want audio. This page publishes the per-model figures for this site, states the configuration each one assumes, and shows how to turn them into a budget.
Credits per generation, by model
Each row is that model's baseline configuration, computed with the same routine the workbench uses. Snapshot taken 2026-08-17; the workbench shows the exact figure for your own settings before you generate.
| Model | Configuration | Credits |
|---|---|---|
| Pollo 1.5 | 480p / 5s | 20 |
| Seedance Pro Fast | 720p / 5s | 35 |
| Pollo 1.6 | 720p / 5s | 40 |
| Sora 2 | 720p / 4s | 50 |
| Seedance 1.0 Lite | 720p / 5s | 50 |
| Google Veo 3.1 Fast | 720p / 4s | 70 |
| Kling O1 Pro | 1080p / 5s | 80 |
| Seedance 1.0 Pro | 720p / 5s | 80 |
| Vidu Q2 Turbo | 720p / 5s | 80 |
| Kling 2.6 | 1080p / 5s | 100 |
| Pika 2.2 | 720p / 5s | 100 |
| Runway Gen-4 Turbo | 720p / 5s | 150 |
| Vidu Q2 Pro | 1080p / 5s | 150 |
| Seedance 2.0 | 720p / 5s | 180 |
| Kling 2.1 Master | master / 5s | 200 |
| Runway Gen-3 Alpha | 720p / 5s | 200 |
| Google Veo 2 | 720p / 5s | 400 |
| Google Veo 3 | 720p / 8s | 670 |
Range at baseline settings: 20–670 credits. A model that looks expensive per generation is not automatically the wrong choice — it is the wrong choice for a draft, and often the right one for the final take.
What an allowance actually buys
Worked with Seedance 1.0 Lite at 720p / 5s (50 credits per generation). Swap in any row above to redo the arithmetic for the model you intend to use.
| Balance | Monthly | Credits | Seedance 1.0 Lite runs |
|---|---|---|---|
| Free starter balance | $0 | 50 | 1 |
| Starter | $3.99 | 300 | 6 |
| Lite | $19.00 | 1,500 | 30 |
| Pro | $49.00 | 4,350 | 87 |
Plan figures come from the pricing page, which stays the source of truth. Runs are whole generations at one configuration; a real month mixes models and settings.
Budget backwards from output, not forwards from plans
The mistake is multiplying finished videos by a plan price. Multiply attempts by configuration cost instead, because attempts are what consume credits.
Know the number before you spend it. The workbench prices the exact model and settings combination in front of you before you submit. A tool that cannot tell you what a generation costs until after it has charged you is not answering a pricing question, it is answering a design question.
Fail cheap. The expensive mistake is not picking a costly model, it is running a costly model on a prompt that was never going to work. Get the shot right at the lowest viable setting — start with a Seedance draft or another low-cost row above — then re-run only the take that worked.
Price a realistic attempt-to-keeper ratio. Three or four attempts per usable result is normal when you are new to prompting, and that is not a failure of the tool. If the arithmetic does not work, the fix is almost never a larger plan — it is a cheaper test loop.
Ready to see it priced live? Open the text to video workbench and the required credits appear before you generate. Comparing specific models first? The Sora 2 page and every other model page carry the same generator.
Cost FAQ
How much does one AI video cost?+
There is no single figure, because the cost of a generation is the model multiplied by the settings you choose. Across the models routed on this site the baseline cost runs from 20 credits (Pollo 1.5) to 670 credits (Google Veo 3). Within one model, raising resolution, extending duration or turning on audio all move the figure. The workbench shows the exact credit cost for your selected combination before you generate.
Why do comparison tables of "price per AI video" disagree with each other?+
Because each table quietly fixes one configuration and compares that. Change the resolution, the duration, or whether audio is on, and the ranking between tools reorders. A price per video is only meaningful next to the configuration it assumes, which is why every row in the table on this page states its configuration.
Do failed generations still cost credits?+
No. When a job reaches a failed state the credits charged for it are returned to your balance, so a failed attempt does not consume your starter balance.
What is the cheapest way to test a prompt?+
Run the idea at the lowest viable setting first — shorter, lower resolution, no audio — and only re-run the take that worked at the quality you actually need. Most of your discarded attempts should be the cheap ones.
Are these credit figures fixed?+
No. They are a snapshot taken on 2026-08-17 using the same calculation the workbench performs, and providers change their rates. Treat the table as a planning aid and the figure shown in the workbench before you generate as authoritative.