RenderDrama

What a second of AI video costs, channel by channel

How many generations may run together

Concurrency is how many generations a channel will run at once for one account. It has no effect on what a second costs and decides how many seconds arrive in an afternoon, which is what a delivery date is made of. As of 2026-10-13.

Four ways this gets answered, and what each is worthOne channel publishes a figure and a monthly price for raising it, which makes throughput a budget item. One publishes a figure per package and one a figure per membership tier. One publishes a ratio with no base, which fails when somebody has to decide whether a tier is enough.A figure and a priceFive sessions free, each furtherone about $2,080 a monthA number and a costThroughput becomes a linein the budgetA figure per packageTwenty concurrency, listed withthe unitsA numberA schedule can be built andchecked against itA comparativeTwice the tier below, withneither quantifiedA ratioNothing can be sized;measure it insteadHow many generations run at onceWhat each answer lets a production do
Fig. 1 Submit a batch the size a real episode needs and time the whole batch. The useful number is finished segments per hour, not latency per segment.
Concurrency, as published. Recorded 2026-10-13.
ChannelWhat is publishedRead from
HeyGenOne tier described as twice the concurrency of the tier below, neither quantifiedThe pricing page
Kling AITwenty concurrency per developer package; unlimited tasks on the membershipThe developer prices and the membership page
SceneMixerThree at once on a free account; five, eight, ten, fifteen and twenty by paid tierThe paid plan guide
ViduFive concurrent sessions free, each further one at about $2,080 a monthThe API price page

Inclusion rule. Channels whose public pages address how many generations may run at the same time. Silence is the norm across this register and is not listed channel by channel. Order. Alphabetical by channel name.

1It is a schedule constraint wearing a pricing page

A rate reachable one job at a time and the same rate reachable twenty at a time are not the same offer. Batch work is where that binds: reference images, then every segment of an episode, then a burst of regenerations after review.

None of it changes the invoice. All of it changes the calendar, which is why a price comparison cannot express the difference in its own units.

2Figures, a price and one ratio

Of the four answers published here, one is a number per package, one is a number plus a monthly price for raising it, one is a number for every membership tier, and one is a ratio with no base. A ratio works on a tier grid and fails as soon as somebody has to decide whether a tier is enough; a number per tier lets the decision be made before buying.

3Why sellers keep it vague

A published concurrency limit converts spare capacity into something buyers will hold a company to. Capacity is exactly what an operator wants to keep discretionary, so the usual answer is silence and the next most usual is an adjective.

4Measuring it takes ten minutes

Submit a batch the size a real episode needs and time the whole batch, not one job. The useful number is finished segments per hour, and it is the figure a schedule should be built on whether or not a page states it.

5It interacts with the ceiling and the retake rate

A production needing sixty generations an episode at a fifteen-second ceiling, with one attempt in three discarded, is ninety calls per episode. At five in flight that is a long afternoon; at one in flight it is a week, and neither number appears on a rate card.

So the three figures to collect together are the ceiling, the rejection rate and the parallel limit. Any one of them alone underestimates how long an episode takes to produce.

6Pages these figures were read from

A definition entry: what the term means, and the figures vendors put in public about it, each figure read on 2026-10-13. Related: Credit allowance, Credit validity.