RenderDrama

What a second of AI video costs, channel by channel

A price printed in money, and what it leaves open

Currency per second is the shape that arrives finished: a figure in dollars set against one second of delivered output, with nothing for this ledger to compute. It is still unusable until the resolution it belongs to is read, because most channels printing one print several. As of 2026-09-12.

How a money figure arrives, and what the row does with itA figure in currency per second reaches this ledger in three conditions: attached to one named tier, repeated across several tiers on the same vendor page, or printed as a band with no tier named. Each condition produces a different row, and none of them is resolved by averaging.What the published figure is attached toOne figure, one named tierStraight onto theledgerThe row carries thefigure, the tier and theaddress it was read from.Several figures, severaltiersOne row for eachQuoting the lowestdescribes a choice madeinside the channel, notthe channel.A band, no tier namedRecorded as a bandA midpoint is a figurenobody is charged and itflatters a wide spread.Three conditions, three kinds of row
Fig. 1 Averaging several published figures into one would invent a price nobody is charged, so a channel with four tiers keeps four rows.
Figure as published on the ledger, by channelEach band runs from the cheapest published figure to the dearest on the same channel. List prices, never quotes.Figure as published on the ledger, by channelWaveSpeed AI$0.01 to $0.15 per…$0.01 to $0.15 per secondRunware$0.025 to $0.12 per…$0.025 to $0.12 per secondVidu$0.035 to $0.075 per…$0.035 to $0.075 per secondfal$0.05 to $0.4 per second$0.05 to $0.4 per secondMiniMax$0.05 to $0.13 per…$0.05 to $0.13 per secondVeo$0.05 to $0.60 per…$0.05 to $0.60 per secondWan$0.05 to $0.20 per…$0.05 to $0.20 per secondKling AI$0.084 to $0.42 per…$0.084 to $0.42 per secondNovita AI$0.084 to $0.168 per…$0.084 to $0.168 per secondLTX Studio$0.09 to $0.39 per…$0.09 to $0.39 per secondReplicate$0.09 to $0.25 per…$0.09 to $0.25 per secondDreamina$0.097 per second$0.097 per secondSora$0.10 to $0.70 per…$0.10 to $0.70 per second
Fig. 2 Each band runs from the cheapest published figure to the dearest on the same channel. List prices, never quotes.
Channels that print the figure in money, and the tier each figure is attached to. Recorded 2026-09-12.
ChannelFigure as publishedTier named beside it
WaveSpeed AI$0.01 to $0.15 per secondNot named; the page says settings move the figure
Runware$0.025 to $0.12 per secondNot named; one figure per model and variant
Vidu$0.035 to $0.075 per second540p, 720p and 1080p
fal$0.05 to $0.4 per secondNot named; one figure per model
MiniMax$0.05 to $0.13 per second480P, 768P and 2K, across two models
Veo$0.05 to $0.60 per second720p, 1080p and 4K, across three model tiers
Wan$0.05 to $0.20 per second480P, 720P and 1080P on one model
Kling AI$0.084 to $0.42 per second720p, 1080p and 4K
Novita AI$0.084 to $0.168 per secondNot named; audio is the step, not resolution
LTX Studio$0.09 to $0.39 per secondNot named; the band covers settings
Replicate$0.09 to $0.25 per second480p and 720p on one model
Dreamina$0.097 per second720p, on one plan billed annually
Sora$0.10 to $0.70 per second720p, 1024p and 1080p, across two models

Inclusion rule. Channels whose own page states a figure in currency against one second of output. A channel that states credits instead belongs to the credit shape even where the conversion is easy. Order. Ascending by the lowest figure the channel publishes; ties break alphabetically.

1Why this shape is kept as its own column

Every line on this ledger ends up as a price for one second of finished picture, and the channels that already print it that way need nothing done to them. Marking those rows is not tidiness. It separates the lines where nobody on this side of the page touched the arithmetic from the lines where somebody did.

The distinction earns its keep when a reader disputes a number. On a converted line the dispute has two possible homes: the vendor's figure, or the conversion applied to it. On a line in this shape there is one, and the argument happens on the vendor's own page rather than here.

2The figure is incomplete without the tier beside it

Vidu prints three of these figures. Veo prints several across three model tiers. Sora prints four before a batch modifier halves all of them. Quoting a channel by whichever is lowest is the commonest way a comparison flatters a product, which is why the tier travels in the row instead of a footnote.

Wan is the plainest illustration here: one model, three resolutions, and the dearest second costs four times the cheapest. Nothing about the model changed between those lines. A budget written against the bottom figure and delivered at the top one is wrong by that factor and by nothing else, and no amount of care with the rate fixes it.

3What a money figure still refuses to say

The rate prices a second the channel considers delivered. It is silent about the seconds discarded on the way there, and exactly one page in this register — Google's — states that a generation which fails is not charged. Everywhere else the question is absent from the price page rather than answered badly.

Two more things sit outside it. How long one generation may run decides how many joins a ninety-second episode needs, and a join is where picture breaks and gets regenerated; that is on clip limits. How many generations may run at once decides whether the rate is reachable the week a deadline lands; that is on parallel generations. Folding either into the headline figure would make it look more complete and leave it less checkable.

4What the seconds buy

Price aside, what goes into the seconds behind “Currency per second” is described in SceneMixer's guides:

Every segment starts with a short header that sets the stage before the first shot: Location (a scene chip), Time, Weather, the Opening frame (what the first frame shows), Set dressing (what is in the room, and any background people) and Props (the objects the segment uses).

SceneMixer, The Segment Header, Field by Field (read 2026-10-13)

Video is 16:9 landscape by default, with 9:16 vertical available; you set it once per project. Segments are typically 4 to 15 seconds, and some model tiers go to 30.

SceneMixer, AI Short Drama Generator — “What length and aspect ratio does it output?” (read 2026-10-13)

5Sources

Every figure above is read from the vendor page linked in the ledger or in the unpriced entries, checked 2026-09-12. What makes a shape convertible at all is on the method page. The other shapes: Credits per second, Plan prices. All four: price shapes.