---
title: "What an AI Film Actually Costs, in Google Flow Credits"
description: "Twenty shots, 3:20, and a real credit ledger: why the floor is 200 credits, why you should plan for 600-800, and what the free tier actually makes."
canonical: "https://www.razi.pro/blog/google-flow-credits-what-an-ai-film-costs"
date: "2026-09-06"
tags: ["Google Flow", "Veo", "AI", "Video Generation", "Costs", "Filmmaking"]
source: "razi.pro"
---

# What an AI Film Actually Costs, in Google Flow Credits

Before you read on, guess: how many credits does it take to generate a finished three-minute animated film?

Every page I could find answering that question is a rate card. They tell you what a credit costs and stop, because none of them has finished anything. So here is a ledger instead, from a film that exists — twenty shots, 3:20, a score, and narration in two languages.

The short version: **the generation is the cheap part, and the number you should plan around is not the one on the pricing page.**

## The floor is 200 credits

The film is twenty shots. Veo generates in clips of up to eight seconds, so twenty clips is the whole picture.

At the time of writing, one Veo 3.1 – Lite generation costs **10 credits**:

| Model | Credits per generation |
|---|---|
| Veo 3.1 – Lite | 10 |
| Veo 3.1 – Fast | 20 |
| Veo 3.1 – Quality | 100 |

Twenty shots on Lite is **200 credits**.

That is the floor, not the bill. It assumes every shot came back right the first time, and no shot comes back right the first time.

> Prices move. This table was checked against Google's own [credit documentation](https://support.google.com/flow/answer/16526234) on 6 September 2026, and it is the first thing to re-check if you are reading this later.

## The number that actually matters is re-rolls

Here is the decision that shaped the entire budget, and it is not the one you expect.

The obvious instinct is to spend credits on quality — generate the important shots at the highest tier. On generative video that instinct is wrong, and expensively so.

A Quality render of a shot where the model misunderstood the action is worth nothing. A Lite render on the fourth attempt, where it finally understood, is worth everything. You are not buying pixels. You are buying attempts.

So: shoot everything on Lite, and spend the difference on re-rolls. At 10 credits a go, a re-roll is cheap enough that you stop rationing them, which is the point — rationing re-rolls is how you end up accepting a shot you do not like.

Budget **two to three re-rolls per shot** and you are planning for 600–800 credits, not 200. That is the honest number.

## The setting that quietly costs you a month

Before you generate anything, check the model selector.

Flow offers Gemini Omni Flash alongside the Veo tiers, and it may not be the one you want. Nothing in the interface flags that you have switched families. Check which model is selected before your first generation, not after your twentieth, because the difference compounds across every clip and every re-roll.

Three other settings are worth thirty seconds each:

- **Confirm before generating: always.** On *never*, an agent run can spend an allocation without asking.
- **Video count: ×1.** ×4 generates four variants per click and bills for four. Re-roll deliberately.
- **Aspect ratio**, before you shoot rather than after.

## Can you do it on the free tier?

Yes, and this is the part the rate cards bury.

The free tier gives **50 credits a day** and includes Veo 3.1. Fifty credits is five Lite generations. So a twenty-shot film is four days of shooting if nothing needs re-rolling, and realistically **ten days to two weeks** once you budget attempts properly.

Two things to know before you plan around it:

**Credits do not roll over.** Not daily, not monthly. Skipping a day does not bank fifty credits for tomorrow — it loses them. If you are shooting on the free tier, shoot a little every day rather than saving up for a weekend.

**Your daily refresh is triggered by your first generation**, not by midnight. Worth knowing if you are counting on a specific reset time.

The paid tiers add a monthly pool on top of the same 50 a day. The smallest of them carries **200 monthly credits** — which, if you have been following the arithmetic, is exactly one film's baseline. You do not need the largest plan to make one of these. Check the [current plans](https://labs.google/flow/about) before assuming otherwise.

One caveat that matters if you are staying free: video upscaling is a paid feature. The free tier includes image upscaling, not video. Plan your delivery resolution around what you can actually export.

## Everything else cost nothing

This is where the ledger stops being about credits, and it is the genuinely surprising half.

| Item | Cost |
|---|---|
| Reference images | free tier, separate pool from Flow credits |
| The score | free — generated locally on a consumer GPU |
| English voice cloning | free — local |
| Malayalam voice cloning | free — local, MIT-licensed |
| Stock-voice narration | free tier |
| Transcription | free tier |
| Assembly, mixing, subtitles, encoding | free — FFmpeg |
| **Marginal cost beyond the video generation** | **zero** |

Every stage after the picture ran on hardware I already owned or on a free tier. The score, both cloned voices, the transcription and the entire assembly added nothing to the bill.

That is not a boast about frugality. It is the actual shape of this kind of project, and it is why the pricing pages mislead: they price the one stage that has a price and imply it is the cost of the thing.

## What the subscription actually bought

Wall-clock. Not capability.

A monthly credit pool collapses the shoot from a two-week drip into one or two sittings. That is worth paying for if your constraint is calendar time. It is worth nothing if your constraint is money, because the free tier makes the same film — the same models, the same resolution, the same twenty shots — just spread across more days.

Nobody selling you a plan is going to say that, so it is worth saying plainly.

## The honest total

If you want one number: **plan for 600–800 credits** for a twenty-shot film, not 200. The floor is a fifth of the real budget.

And if you want the number that actually hurt, it was not credits at all. It was the reference-image iteration, one long debugging session on narration duration, and the narration edit. Video generation was the fastest, cheapest, most predictable part of the whole project.

The tools have got cheap. The taste has not.

---

*This is part of a longer production log — [everything that broke making the film](https://www.razi.pro/blog/making-an-ai-animated-film-with-veo-and-flow), including the shot that came back as an astronaut and the narrator that invented a language. There is also a [making-of video](https://razi.pro/flow-making) if you would rather watch than read.*