What One Finished AI Video Actually Costs Me (Real Numbers)

Every AI video tool prices the same thing: the render. That is the cheapest part of a finished video, and it is the only part anyone talks about.

I run a daily pipeline — three vertical clips a day for a clothing store, plus the same clips relocalised into six more country accounts. That is around twenty-one published posts a day off three renders. Because of that I have a fairly exact picture of where the money actually goes, and it is not where the pricing pages point.

This is the breakdown of one finished clip: what the render costs, what the four steps after the render cost, and the two numbers that decide whether any of it is worth doing.

The render is roughly a tenth of the work

A "finished" clip in my pipeline is not what comes out of the model. It is the model output plus: a voiceover, burned captions, a music and ambience bed, a two-second brand end card, and a written caption with a product link. Then it gets scheduled to four platforms.

The generator produces exactly one of those seven things.

That matters commercially because the render is the only step with a per-unit price attached. The other six are either a fixed monthly cost or your own time, so they vanish from the comparison — and then people conclude the expensive tool is the expensive part of their setup. It usually is not.

What the render actually costs

Credit-based platforms make this straightforward to compute. On the plan I use, a 15-second vertical clip at 720p costs 37.5 credits on the mini-tier image-to-video model. Reference images for that clip cost about 2 credits each at 2K on a top-tier image model.

So one clip, generated properly with its own reference stills:

The rejected render is the line people leave out. My rejection rate on first attempts has never been zero — wrong garment, a hand that goes wrong, text the model garbled, a start frame that never showed the product. If you plan at the sticker price you will be over budget by month's end every month.

The rest of the chain, honestly priced

Voiceover. A fixed monthly text-to-speech subscription, not per clip. At my volume the per-clip cost rounds to a couple of cents. This is the single best value step in the whole pipeline and the one beginners skip.

Captions. Free. Word timestamps from a free-tier speech API, burned in with ffmpeg. The only cost is the hour it took to build the script once.

Music and ambience. Free — one licensed instrumental I own, mixed under every clip with a consistent level and a swell at the end. One track reused forever beats a different trending sound per clip for a brand feed.

End card. Free. One PNG, two seconds, concatenated on.

Scheduling and publishing. Free via the platform APIs, and this is where the real cost hides: it is where things silently fail. A post that reports success and never appears costs you the whole clip.

The number that actually matters: cost per published post

Here is the leverage, and it is the reason I stopped optimising the render price.

One render, once finished, gets translated into six more languages — script, voiceover, on-screen cards and caption — with the same footage. Each locale gets its own account, its own slot and its own day, because identical content across sibling accounts on one IP is a spam signal, not a growth hack.

Same footage, twenty-eight ways. The translation step costs nothing but compute I already pay for, and the voiceover is inside the subscription. Nothing about that requires a better model — it requires a script that runs after the render.

This is the whole argument for building a pipeline instead of buying more renders. Going from one output per render to twenty-eight cut my cost per post by more than an order of magnitude, and no tool change would have come close.

What the cost per post has to beat

Cost only means something against a return, so here are the two numbers I judge it against.

On the organic side: over 30 days on one channel, 90 clips produced 37,306 views, and the top 27% of clips carried over two-thirds of them. The median clip did well under a hundred views. You cannot pick the winner in advance, so the only strategy that works is volume — which means cost per post is the metric, not cost per render.

On the paid side: the ads built from these clips are currently buying signups for a $9/month subscription at about 7.94 EUR per purchase across 43 purchases in seven days. That is under one month's subscription price, which is the threshold I keep. Note that a subscription makes ROAS useless — the platform records only month one, so reported ROAS can barely clear 1.0 however well the ad works. Cost per purchase is the honest measure. If you sell one-off products, ROAS is fine.

Where the money actually leaks

Three leaks, all bigger than the difference between any two generators:

Unused renders. Clips finished and never published are a total loss, and they accumulate quietly. I once found fourteen finished autumn clips on disk that could never be posted — they had a price burned into an on-screen card, and prices change, so the pixels were unfixable and the raw renders were gone. Fourteen renders, roughly 1,100 credits, zero posts. Never burn a price into a frame, and never delete the raw render.

Renders you cannot post soon. If you can publish three clips a day and you generate ten, seven are going into a queue that will make them stale before their slot. Generate to your posting capacity, not to your credit balance.

Expiring credits. Subscription credits usually do not roll over. Unspent credits at renewal are a pure write-off, and worth a calendar reminder — spend them down deliberately in the last days of the cycle rather than discovering the loss afterwards.

So what does one video cost?

Straight answer, from my own accounts: about 80 credits per finished master, and under 3 credits per published post once localisation is in place. On a mid-tier subscription that is a few cents per post.

The render is not the expensive part. Not publishing is.

If you are comparing tools on price, the useful comparison list is here: Higgsfield pricing, Kling pricing, Runway pricing, Seedance pricing and Veo 3 pricing. But before you switch, work out your cost per published post on what you already pay for. Most people find the answer is a pipeline problem, not a pricing one.

Related: how many AI videos per day, AI video for multiple markets and the posting slot that beat my creative by 20x.

Adobe prices Firefly video in generative credits rather than clips, which is why the plan page never answers the question you arrived with. Video is the most credit-hungry thing an Adobe subscription can spend on, so the headline price tells you almost nothing — the number that matters is the monthly allowance divided by the credit cost of one generation at the length and resolution you actually ship. On raw quality per credit Firefly is mid-tier and loses to the current leaders; what it sells instead is commercial indemnification and living natively inside Premiere. If neither of those is your constraint, the money goes further elsewhere. The arithmetic is in Adobe Firefly video pricing.

I publish the actual scripts, the credit ledger and the failures inside the community. If you want the working pipeline rather than the write-up, join us here.

Looking for something else? Browse all 124 AI video guides in one list.