AI Video Generation Cost: What Each Clip Actually Costs

AI video generation cost per second, by model

On RelayMe, AI video generation is billed per second of output. The rate depends on the model, the resolution tier and, for some models, whether an audio track is generated. The table below is built from the live model catalogue each time this site is published, in credits per second; 1,000 credits is about US$1.

A clip’s video fee is that rate multiplied by a duration the model supports. Veo 3.1 and Veo 3.1 Fast are cheaper without audio, Grok Imagine costs the same either way, and MiniMax-H3 always includes an audio track. Image inputs, text/LLM steps and retries are charged separately.

What one usable clip actually costs

Here is one dated run from our own production account on 25 August 2026. A 1:1 product image generated with Nano Banana on the sale channel cost 20 credits and took about half a minute. That image became the first frame of a Veo 3.1 Fast image-to-video clip — 4 seconds, 720p, 16:9, with audio — which cost 600 credits and took about a minute. The whole chain came to 620 credits, about US$0.62.

That is a record of one run, not a quote: rates change, and the table above shows today’s. It covers a single attempt that we accepted. Text/LLM steps, retries and editing time are billed or tracked separately.

To budget, divide the total credits charged for a batch by the clips that pass review. If only half of your attempts pass, the cost per accepted clip doubles: 620 credits per attempt becomes 1,240 credits per usable clip.

Check the current pricing and credit rules before selecting a model; a dated example is not a current quote.

Once a still image is approved, the image-to-video guide explains the next stage and its dated worked example.

What actually makes up AI video ad cost

The real cost of an AI video ad has four parts: model fees (usually tied to duration and quality tier), re-run waste (regenerating unsatisfying results), human time (prompting, screening, editing), and exploration cost (the batches it takes to find a usable style). The listed model price is only the first of them.

The first step to controlling spend is tracking these four buckets separately so you can see where the optimization headroom is. RelayMe’s current rates are in the table above and on the pricing page; confirm them before a large batch.

Compare models at the same duration

Compare models at a duration every candidate supports. When this guide was reviewed, Veo 3.1 and Veo 3.1 Fast offered 4, 6 or 8 seconds, Grok Imagine a set of fixed lengths up to 15 seconds, and MiniMax-H3 5 to 15 seconds, so a 4-second round cannot include MiniMax-H3 while 6 or 8 seconds can. Record each model’s actual resolution too: 720p and MiniMax-H3’s 768P are not the same setting.

Use price and supported settings to build a shortlist, then compare your own outputs. A higher price does not establish higher quality for every shot, and a lower per-second rate does not establish a lower cost per usable clip. Keep the prompt, reference input and review criteria stable; record duration, resolution, audio, retries and actual task spend for each result.

Check Veo 3.1 vs Veo 3.1 Fast and Grok Imagine vs MiniMax-H3 for the inputs, duration settings and prices of each model.

How mainstream models price generation

Billing for mainstream video models such as Veo and Grok Imagine typically depends on three factors: clip duration, resolution tier, and generation mode (fast draft versus high quality). The same prompt can cost several times more after switching tiers.

Do not compare list prices alone — estimate by your real usage: iterate on low tiers during A/B testing, then render final cuts on high tiers. Since prices and tiers shift with releases, note the date of each rate in your budget sheet.

For a concrete anchor, use the 25 August 2026 run above: at that time Veo 3.1 Fast billed 150 credits per second with audio, so the 4-second clip cost 600 credits, and the seed image added 20.

Four cost-saving tactics that work

First, image before video: lock composition and style with low-cost image generation, then render video only after approval, cutting expensive re-runs. Second, template your prompts: save proven prompts as reusable templates and swap only the product variables for new items.

Third, split tiers: drafts on the fast cheap tier, finals on the high-quality tier. Fourth, track re-run rate: measure re-runs per asset type and fix high re-run prompts before paying for a pricier model. Check the effect of each change in task history rather than assuming a saving.

Audit spend with task history

RelayMe keeps a task history of every generation, which lets you audit spend against output: which workflow has the highest usable rate, and which prompt patterns waste the most.

Review it weekly and build a simple report of clips generated, clips used, and re-runs. When an asset type stays low on usable rate, revisit the prompt and reference material before changing models or tiers.

Our own August 2026 run shows what that audit looks like line by line: 20 credits for the seed product image and 600 credits for the 4-second clip. That per-step record is what a weekly review needs.

Scale your video production workflow

Once a single flow is stable, freeze it as a RelayMe workflow template: a language model step expands product details into video prompts automatically, and the video step renders the result, so operators only maintain the product sheet.

At scale, the lever is no longer unit price but usable rate and screening time — both usually outweigh the price gap between models. Validate template stability on small batches before ramping volume.

At scale, the per-second rate becomes a planning number. At the rate of the 25 August 2026 run (600 credits for a 4-second Veo 3.1 Fast clip with audio), 100 clips budget at 60,000 credits before retries and image inputs. Compare that with the monthly credits of each plan on the pricing page, or cover it with top-up credits, which do not expire.