Pricing pages in AI video are written to be hard to read. Credits that do not map to money, packs with bonus tiers, "from" prices that apply to plans nobody buys, and pricing pages that return a 404. I run a company in this market, so I have both the incentive to complicate things and the data to explain them. Here is the explanation.

The three cost models

Per-minute. You buy minutes of finished video a month. Simple to understand, brutal in practice, because a thirty-second ad that needs four takes costs two minutes, not thirty seconds. Presenter-style avatar tools such as HeyGen and Synthesia tend to use this model.

Credits with a hidden exchange rate. You buy credits, and a video costs a number of credits that depends on length, model and resolution. This is the most common model (Creatify, Captions, MakeUGC) and the least transparent, because the exchange rate to money is never on the page. The real cost per finished ad, after the takes you discard, is often three to five times the headline number.

Pricing on request. No public price at all. Arcads works this way; third-party breakdowns put it at roughly $110 to $550 a month. It means the price is whatever the sales call decides.

What the tools cost, as of September 2026

Prices read from each vendor's pricing page in September 2026; they change, so check before you buy.

What "one video" costs when you count the takes

Every vendor's marketing number is the cost of the take you keep. The number that matters is the cost of the take you keep plus the takes you throw away. In our data and in what other tools' users report, expect to discard between a third and a half of first outputs: a script rewrite, an avatar swap, a length change. Whatever the per-video figure is, plan for roughly double.

The model we chose, and why

We priced at parity: one credit is one US cent, on every plan, and a plan always grants exactly the credits it costs. A $19.99 plan is 1,999 credits. Extra credits cost the same rate, with no bonus tiers. And the exact credit cost of any generation is shown before it runs, so the "will this be expensive" question is answered before the money moves.

It removed our most common support ticket, which was some version of "what does this cost". It also removed a lever most of our competitors use, because there is nowhere to hide a margin in a system where everyone can do the arithmetic. We think the trade is right, but it is a trade.

How to compare tools honestly

Ask four questions of any AI video pricing page:

1. What is the exchange rate between credits and money, in writing?
2. Is the price of a generation shown before it runs?
3. Do unused credits roll over, and to what cap?
4. What happens to the credits when a generation fails?

If a page cannot answer all four, the real price is higher than the headline.

A worked example

Say you want ten thirty-second UGC-style ads a month and you expect to keep one take in two. That is twenty renders. On a per-minute plan with ten minutes, that is your entire allowance with no room for anything else. On an opaque credit plan, you cannot know until you try. On a parity plan, you multiply the shown cost by twenty and you have the number before you start.

This is written by someone with a product to sell, so check the arithmetic yourself; the point is that you should be able to.

Ionel Doboaca is co-founder and CTO of ACE AI Spark (aceaispark.com), an AI UGC platform that makes creator-style ads and social video and publishes them to six networks. Its openly-AI influencer, Jeyla Shore, has 800K+ followers.