How to budget an AI short film (worked example)
Data checked: October 1, 2026
A multi-scene project is cheaper to plan scene by scene than with one average rate. Different scenes call for different models, and the number of takes varies by how hard the shot is. In this example, a 94-second film with five scenes costs $62.59 before contingency and $71.98 with 15% contingency.
The scene models and take counts below are illustrative choices, not recommendations. Prices come from our dataset and are per second of generated video, without audio.
Try the calculator: AI Video Budget Planner →
Scene-by-scene budget
| Scene | Model | Resolution | Seconds × takes | Cost |
|---|---|---|---|---|
| Opening establishing shot | Google Veo 3.1 Fast | 1080p | 12 s × 3 | $4.32 |
| Character close-ups | Kling 3.0 Pro | 1080p | 24 s × 4 | $10.75 |
| Chase sequence | Luma Ray 3.2 | 1080p | 30 s × 3 | $21.60 |
| Dialogue scene | Google Veo 3.1 | 1080p | 20 s × 3 | $24.00 |
| Closing shot | Runway Gen-4.5 | 720p | 8 s × 2 | $1.92 |
| Subtotal | $62.59 | |||
| Contingency (15%) | $9.39 | |||
| Total | 94 s final | $71.98 |
What the number means
The total works out to about $45.95 per finished minute. Most of the spend is concentrated in the scenes with the most generated seconds, so reducing takes on those scenes saves more than switching models on short ones.
Contingency covers shots that fail repeatedly, scope changes and the occasional re-render. Fifteen percent is a starting point; increase it if your concept relies on hard-to-control motion or consistent characters.
Costs this budget leaves out
Music, voice, editing software, upscaling and your own time are not included. Native audio may also be priced separately by some providers.
Frequently asked questions
How much does an AI short film cost?
In our example, 94 seconds of final footage costs $71.98 including contingency. Your number depends on models, resolution and takes.
How many takes should I budget per scene?
It varies with shot complexity. Two to four is a common planning range, but measure your own retake rate on a test scene.
Why add contingency?
Because generation is not deterministic: some shots need many more attempts than planned.
Sources
Prices updated: October 1, 2026
- Google Gemini API pricing (Veo) ↗
- OpenAI API pricing (Sora) ↗
- Runway API pricing ↗
- Runway plans ↗
- Kling developer pricing ↗
- Luma API pricing ↗
- Luma plans ↗
- Pika pricing ↗
- Higgsfield pricing ↗
Prices are collected from public pages and may change without notice. Entries marked 'secondary' were read through a third-party aggregator because the official page did not render to our fetcher; the official URL is still listed as the source of record.
- Higgsfield and Pika publish credit-based subscription pricing; we did not find public per-second API rates for them, so they appear as effective subscription rates.
- Veo is priced from the Gemini API (Veo on Vertex AI or inside Google AI plans may differ).
- Taxes, free tiers, batch discounts, and enterprise contracts are not included.