Every AI video tool publishes a price. Almost none of them publish the number you need, which is: what does it cost to get one ad I would actually run?
Those are very different figures, and the gap between them is where budgets disappear.
Why the sticker price misleads
Three separate reasons, and most pricing pages exploit all three.
1. Credits are not outputs
"500 credits for $29" is meaningless until you know what a credit buys. Credits typically get consumed by:
- Image generation, often per image rather than per set
- Video generation, usually per second
- A resolution multiplier, where 1080p can cost several times 480p
- Regeneration, which may or may not be charged
A tool charging per second of video at 1080p can easily be five times the effective cost of one charging per clip, at the same headline credit price.
2. The first output is not the output
Nobody runs the first generation. The realistic path is three to six attempts — a frame that was not right, a script that missed, a shot that drifted.
If each attempt costs full price, your real cost per usable ad is four or five times the quoted cost per generation. A tool that charges for regeneration is substantially more expensive than one that does not, regardless of the headline rate.
3. Monthly credits usually expire
Unused credits often do not roll over. If you buy a tier for the occasional busy month, you pay for that tier every month. The effective annual cost is the subscription, not the usage.
The number to calculate
Cost per usable creative = (total spend over a period) ÷ (ads you actually ran from it)
Not generations. Not credits. Ads you ran.
Work it out over your first month, honestly including the failures. Most people find it is three to six times what they assumed from the pricing page.
Doing it for a real workflow
Take a typical four-shot vertical ad and price each stage. The shape matters more than any specific vendor's numbers:
| Stage | Typically charged by | Attempts in practice | |---|---|---| | Product analysis / script | Free or trivial | 1–2 | | First frames (4 shots) | Per image | 2–3 sets | | Video generation (4 clips) | Per second × resolution | 1–2 | | Assembly | Usually free | — |
The thing to notice: frames are cheap and video is expensive, usually by a large multiple. Which means the single biggest lever on your cost is where you iterate.
Iterating on frames and only generating video once they are right can cut the cost of a finished ad by more than half. Iterating by regenerating whole videos is the expensive habit, and tools that do not let you fix one shot in isolation push you into it structurally.
Resolution is a bigger lever than it looks
Video cost scales roughly with pixel count. Relative to 720p, 1080p is somewhere around twice the cost and 480p around half.
For a vertical feed ad watched on a phone, 720p is almost always sufficient — the difference against 1080p is invisible at that size and viewing distance, and the platform re-encodes your upload anyway.
Which means: testing at 480p and only finishing winners at higher resolution can cut testing costs by three quarters. If you are producing fifteen variants to find one that works, fourteen of them never needed to be sharp.
Length is the other one
Cost is usually linear in duration. An eight-second clip costs twice a four-second one.
Four seconds is enough for most single shots in a feed ad. The instinct toward longer clips is usually a mistake creatively and doubles the bill.
Comparing against the alternatives
Context for whether any of this is good value:
Commissioning a creator. Rates vary enormously by audience size and category, and almost nobody publishes them — treat any quoted figure, including from vendors making a comparison, as unverified. What is structurally true: it is priced per asset, it does not scale down for a long catalogue, and turnaround is days to weeks.
Filming yourself. Near zero marginal cost, bounded entirely by your own time. Perfectly viable for one product; impossible for two hundred.
A production shoot. The highest quality ceiling and the highest fixed cost. Makes sense for a flagship asset meant to last.
Generated video is not cheaper than all of these in every case. Its distinct property is that the cost per additional variant is nearly flat — the fifteenth variant costs about what the first did. For creative testing, where you need volume rather than a ceiling, that is the whole argument. For one hero asset, it is not an argument at all.
Questions that reveal the real price
Ask before subscribing:
- What exactly consumes a credit? Per image, per second, per resolution tier?
- Does regeneration cost full price?
- Do unused credits roll over?
- Is there a watermark on any paid tier?
- What happens when a generation fails? The answer should be an automatic refund.
- Can I generate anything without entering a card? If not, you are buying blind.
A tool that answers all six clearly on its pricing page is telling you something about how it expects to be judged.
A worked habit
If you want one rule that reliably lowers the bill:
Separate the cheap step from the expensive step, and iterate only on the cheap one.
Get the still frame right — composition, product fidelity, the label still legible — while each attempt costs a fraction of a clip. Only then pay for motion. This single discipline matters more than which vendor you choose.
Where Faysell lands
For transparency, since this article is on its blog: Faysell charges one credit per generated image and bills video by the second with a resolution multiplier, so 480p costs half of 720p and 1080p a little over twice. Failed generations are refunded automatically. New accounts get credits to generate a full set of first frames without entering a card, which exists precisely so you can run the arithmetic above on your own product before paying.
The frame and video steps are deliberately separate, for the reason in the previous section. The pricing page has the current rates.
The summary
The number that matters is cost per ad you would actually run, which is several times cost per generation. Frames are cheap, video is expensive, resolution and length are the big multipliers, and where you choose to iterate matters more than which tool you buy.
Test at low resolution. Iterate on stills. Finish only the winners.