AI ads, UGC creators, or an agency: what each costs and what you get

01What does an AI ad actually cost per video?

Nobody publishes that number, the tools included. What they publish is a monthly subscription and a credit allowance. Creatify's Starter plan is $39 a month for 100 credits, and the pricing page does not say what one video costs in credits. HeyGen gives you 600 credits for $29. Without the exchange rate, there is nothing to divide.

So run the arithmetic from the other end. Take the subscription, divide it by the number of ads you would genuinely put in front of a customer this month, and you have a real per-asset figure. It moves with your own output instead of with the vendor's plan names. A $99 plan across ten shipped ads is about ten dollars an ad. Across two, it is fifty, and the plan was never the thing that was expensive.

The shape of that table is the finding. Tools price in credits, a currency they also control. Creator platforms price in dollars but keep the creator's own rate behind a marketplace fee. And two of the best known names, Arcads and Billo, published no price at all on the day we looked.

02Cost per finished asset is not cost per usable asset

Here is the number this market does not print. In our own production log, 57 out of every 100 films that clear every automated check we run still get killed on the second watch, and only 43 survive. Not for a fault a checker can catch. The frame is clean, the type is legible, the physics hold, and the film is dead. We build a surplus of at least two to one for that reason, and anything scoring under 85 out of 100 is killed rather than repaired.

A price per asset is only a price per asset if you would run the asset.

The line we use in every scoping call

What our 43% keep rate does to a $100 video

Apply our own measured rate to a sticker price and the routes separate immediately. A $100 creator video at a 43% keep rate costs $233 for every one you would actually run, because you paid for the ones you binned. The same rate applied to a flat tool subscription changes nothing on the invoice: you burn credits and an afternoon instead. A retainer behaves differently again, because the misses are supposed to happen inside the fee. That asymmetry, not the headline price, is what you are choosing between.

Drag the keep rate down to 20% and watch which column moves. The creator line more than doubles. The tool line does not move at all. That single behavior explains most of the arguments people have about this choice, and it is arithmetic rather than taste. If you want the version of this with real render failures counted in, we broke it out in what failed renders actually cost.

03Who owns the file, and what happens when it fails?

Three different answers, and it is the part of the deal people skip. A tool gives you output under a license attached to your plan. A creator gives you a video licensed for stated channels and a stated period unless you pay to buy it out. A studio should assign you the finished file outright. Ask all three in writing, before money moves, and read what comes back.

Failure divides the same way. A missed generation costs you credits and the hour you spent choosing. A creator video that misses has already been paid for, so the fix is a new order at the same price. A studio miss is supposed to sit inside the fee, which is most of what a fee is for. That is the honest case for hiring one, and it is narrower than agencies like to make it sound.

04What each route is genuinely best at

Every comparison on the internet is written by somebody selling one of the columns, this one included. So here are the same facts arranged to be useful rather than flattering.

05Why does nobody publish a cost per usable asset?

Because it is the number that makes the sale harder. A tool publishing $39 a month is selling a floor. A platform publishing $100 a video is selling a unit that arrives finished. Both are true. Neither includes what happens when the thing lands in your inbox and you would not run it, and publishing that means publishing your own failure count.

We publish ours because a buyer who knows our keep rate can price us properly, and a buyer who does not will work it out by month two anyway. So the question to put to all three routes is not what a video costs. It is what a video you would run costs.

06So which one should you actually pick?

Volume decides it more than category does. Under about six usable ads a month, a subscription tool plus your own taste beats everything, because a retainer divided by four is a bad number no matter how good the retainer is. Above roughly fifteen, the hours you spend rejecting things start to cost more than paying somebody else to do it. That crossover is a judgment rather than a measurement, and it is ours.

The creator route is not a volume play and should not be priced as one. Hire a person when the idea requires a person, which is a specific list rather than a feeling. We wrote it out in when you should still hire a human creator, including the cases where we tell people not to hire us. Whether your product suits synthetic footage at all is a separate question, and it turns on what you are asking the camera to do: that one is in what kinds of products AI UGC actually works for.

One thing survives all of this arithmetic. Every route is priced on what gets made, and you are buying what gets shown. Whoever you hire, the question that should decide it is how many things they were willing to throw away on your behalf, and whether they will tell you the number.