How Much Does AI Video Generation Cost per Second in 2026?

From 3.5 tokens to 75 tokens per generated second depending on the model and resolution: the full price table as billed in SociaLover, three worked examples and the habits that cut the bill without cutting quality.

SociaLover Team · Updated · 14 min read

A second of AI generated video costs between 3.5 and 75 tokens in SociaLover, which bills pay as you go at 1 token = $0.01: from $0.035 per second on Seedance 1.5 Pro in 720p up to $0.75 per second on Seedance 2.5 in 1080p. The model and the resolution set the price, and audio can double it.

What drives the price of a generated second?

Before the levers, the billing frame. SociaLover bills video generation pay as you go: there is no seat subscription, you buy tokens and each generated second consumes them at the model's rate, with 1 token worth $0.01. That makes the cost of a campaign a pure function of what you render, which is why this page is organized around the price of one second: every budget below is that unit multiplied by a duration and a number of takes.

Three variables set the bill: the model you pick, the resolution you render at, and whether the clip carries audio. The model is the biggest lever, because every provider ships several tiers of the same engine. Veo 3.1 bills 53 tokens per second ($0.53/s) while Veo 3.1 Lite bills 7 tokens per second ($0.07/s) in 720p, more than seven times less for the same prompt box and the same duration slider. Kling runs a similar ladder: Kling 3.0 and Kling 3.0 Turbo both bill 19 tokens per second ($0.19/s) with audio included, while Kling 2.6 renders a silent clip at 9 tokens per second ($0.09/s). All figures here and below are the prices as billed in SociaLover, pay as you go, where 1 token equals $0.01; they are not the rates the providers publish for their own APIs.

Resolution is the lever people underestimate. On Seedance 2.5 the same second costs 14 tokens in 480p, 31 tokens in 720p and 75 tokens in 1080p, so the top resolution multiplies the price by more than five compared with the draft one, and by almost two and a half compared with 720p. Seedance 2.0 follows the same curve at 9, 20 and 49 tokens per second. Other models price the jump gently: Veo 3.1 Fast moves from 13 to 16 tokens per second between 720p and 1080p, three tokens more. Knowing which curve your model sits on decides where you can afford to iterate.

One model, three prices: Seedance 2.5 by resolution
480p
14 tk ($0.14)
720p
31 tk ($0.31)
1080p
75 tk ($0.75)
Tokens per second for the same model, Seedance 2.5, at its three output resolutions, as billed in SociaLover, pay as you go (1 token = $0.01). The step from 480p to 1080p multiplies the cost of every second by more than five, which is why the resolution you draft at matters more than almost any other choice once you start iterating.

Audio is the third lever, and the least uniform. Seedance 1.5 Pro doubles with it, from 3.5 to 7 tokens per second in 720p and from 8 to 15 in 1080p. Kling 2.6 jumps from 9 to 19 tokens per second when audio is on, more than double. Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite fold native audio into their rate, as do Kling 3.0 and Happy Horse 1.1, and Seedance 2.5 treats it as an option. One practical caveat: Kling's native audio only speaks English or Chinese, and any other language is translated into English, so a French or Spanish ad is usually generated silent and voiced in the edit. The live rate card for every model sits on the SociaLover pricing page.

The full price table

The table below lists every video model available in Studio Video with its price per generated second at each resolution, in tokens and in converted dollars, as billed in SociaLover, pay as you go. A cell marked not offered means the model does not output that resolution. Four models bill a flat rate per second whatever the resolution, Veo 3.1, Kling 3.0, Kling 3.0 Turbo and Kling 2.6, which is why the same figure appears in both of their resolution columns.

Price per second of generated video in October 2026, in tokens with the exact dollar conversion in parentheses, as billed in SociaLover, pay as you go, at 1 token = $0.01. These are platform prices, not the providers' API list prices. Not offered means that resolution is not available for that model; Veo 3.1, Kling 3.0, Kling 3.0 Turbo and Kling 2.6 bill the same flat rate per second at every resolution they output.
Model480p720p1080p
Veo 3.1 (native audio)not offered53 tk ($0.53)53 tk ($0.53)
Veo 3.1 Fast (native audio)not offered13 tk ($0.13)16 tk ($0.16)
Veo 3.1 Lite (native audio)not offered7 tk ($0.07)11 tk ($0.11)
Kling 3.0 (audio included)not offered19 tk ($0.19)19 tk ($0.19)
Kling 3.0 Turbo (audio included)not offered19 tk ($0.19)19 tk ($0.19)
Kling 2.6not offered9 tk ($0.09), 19 tk ($0.19) with audio9 tk ($0.09), 19 tk ($0.19) with audio
Seedance 2.5 (up to 30 s per clip)14 tk ($0.14)31 tk ($0.31)75 tk ($0.75)
Seedance 2.09 tk ($0.09)20 tk ($0.20)49 tk ($0.49)
Seedance 2.0 Fast8 tk ($0.08)16 tk ($0.16)not offered
Seedance 1.5 Pronot offered3.5 tk ($0.035), 7 tk ($0.07) with audio8 tk ($0.08), 15 tk ($0.15) with audio
Wan 2.6 / 2.7not offered13 tk ($0.13)20 tk ($0.20)
Happy Horse 1.1 (native audio)not offered19 tk ($0.19)24 tk ($0.24)
Grok Imagine 1.511 tk ($0.11)19 tk ($0.19)33 tk ($0.33)

One name is deliberately missing. Sora 2 billed 13 tokens per second in 720p, and Sora 2 Pro between 40 and 93 tokens per second, until OpenAI retired the model: the shutdown was announced on March 24, 2026, the app closed on April 26, 2026, and the API was cut on September 24, 2026. A price table that still quotes Sora 2 in late 2026 is quoting a product you cannot buy. If a workflow was costed around Sora 2's former 13 tokens per second, the closest per second equivalents in the table are Veo 3.1 Fast and Wan 2.6 or 2.7, both at 13 tokens per second in 720p; the quality trade offs are a separate question, covered in our model comparison.

1. Shutdown announced
March 24, 2026
2. Sora app closed
April 26, 2026
3. API cut off
September 24, 2026

The six months it took Sora 2 to leave the market. The API cutoff on September 24, 2026 is the date that matters for advertisers: any pipeline still pointed at Sora 2 after that day simply stopped rendering, which is why the table above no longer carries a Sora row.

Two reading notes. Seedance 2.5 is the only model in the table that renders up to 30 seconds in one generation, where the others produce short clips, and it can return the last image of a clip so the next one starts exactly where the previous one ended. And the dollar figures are not a second price list: each one is the token figure multiplied by $0.01, nothing else, so a model quoted at 20 tokens per second always means $0.20 per generated second.

Read column by column, the spread is wide enough to change how you brief. In 480p the table runs from 8 tokens per second on Seedance 2.0 Fast to 14 on Seedance 2.5. In 720p it runs from 3.5 tokens per second on Seedance 1.5 Pro to 53 on Veo 3.1, a factor of more than fifteen for the same duration. In 1080p it runs from 8 tokens per second, again on Seedance 1.5 Pro, to 75 on Seedance 2.5. The same 10 second brief can therefore bill 35 tokens ($0.35) or 750 tokens ($7.50) depending on where you point it, before any difference in quality enters the discussion.

What does a full ad actually cost?

Per second rates only become concrete once you multiply them by a real brief. Here are three jobs we actually run, with the arithmetic written out, all as billed in SociaLover, pay as you go. None of them needs anything beyond multiplication, which is the quiet virtue of per second billing: the quote and the invoice are the same number.

First, a 10 second UGC style clip on Seedance 2.0 Fast in 720p. The model bills 16 tokens per second, so 10 seconds come to 160 tokens, exactly $1.60 for the take. That tier is quick enough to try several hooks in one sitting, which is the whole logic of UGC video without an actor: volume of takes rather than one precious take.

Second, a series of three episodes of 8 seconds each on Seedance 2.5 in 720p. At 31 tokens per second, each episode bills 248 tokens ($2.48), and the set of three bills 744 tokens, $7.44 in total. Seedance 2.5 hands back the last frame of a clip so the next episode can start from it, which is the mechanic behind Series. One constraint is worth knowing: the provider refuses a photo of a real person as a start image, but accepts images generated inside its own ecosystem. That is why the 30 ready creators added to UGC Creators in October 2026, fictional people delivered as finished UGC images, are Seedance compatible by construction, and every episode animates that exact image.

Third, a 6 second hero spot on Veo 3.1 standard. At 53 tokens per second, the spot bills 318 tokens, $3.18, with native audio already folded into the rate, so there is no separate line for sound.

Three finished ads, in billed tokens
10 s UGC clip, Seedance 2.0 Fast 720p
160 tk ($1.60)
3 episodes of 8 s, Seedance 2.5 720p
744 tk ($7.44)
6 s spot, Veo 3.1 with audio
318 tk ($3.18)
Total tokens billed for the three worked examples, at the per second rates from the table above, as billed in SociaLover, pay as you go. Each figure covers a single take of each clip: a second take of the same clip bills the same amount again, which is why the next section is about iterating at a cheaper resolution before paying the full rate.

Another way to hold the scale in mind is to flip it: $10 is 1000 tokens, as billed in SociaLover, pay as you go. At 16 tokens per second, Seedance 2.0 Fast in 720p, that buys 62 full seconds of footage, six of the UGC clips above with tokens to spare. At 53 tokens per second on Veo 3.1 it buys 18 full seconds, three of the hero spots. At 75 tokens per second, Seedance 2.5 in 1080p, it buys 13 full seconds. The same ten dollars lands anywhere between one finished clip and six, depending on the tier you point it at.

The honest caveat is that these are single take prices. A usable second costs more than a generated second, because the keeper is rarely take one: on the accounts we run, a finished ad usually sits on top of a few discarded takes. That multiplier, not the per second rate, is the real budget line, and it is exactly what the next section compresses.

How to spend less per usable second?

Iterate at a resolution you will throw away, then finalize once at the one you ship. On Seedance 2.5, a 10 second draft costs 140 tokens ($1.40) in 480p and 750 tokens ($7.50) in 1080p, so five full drafts in 480p, 700 tokens in all, still bill less than one single 1080p master. Lock the motion, the framing and the timing at 14 tokens per second, then spend 75 tokens per second exactly once.

Keep clips as short as the idea. Billing is strictly per second, so a 6 second cut of a 10 second idea on Veo 3.1 bills 318 tokens instead of 530, and that 212 token gap ($2.12) repeats on every take you discard along the way. Short is also how hooks behave in feed placements, so the cheaper cut is very often the better ad.

Finally, match the model to the job instead of defaulting to the premium tier. A motion test runs on Seedance 1.5 Pro at 3.5 tokens per second, which makes a 10 second silent try cost 35 tokens ($0.35), or on Veo 3.1 Lite at 7 tokens per second. Background b-roll sits comfortably on Wan 2.6 or 2.7 at 13 tokens per second in 720p, or on Grok Imagine 1.5 at 11 tokens per second in 480p. Keep Veo 3.1 at 53 tokens per second, or Seedance 2.5 in 1080p at 75, for the clip that actually ships. And do not pay for audio you will mute: Kling 2.6 drops from 19 to 9 tokens per second without it, and most ads replace native sound with a voiceover and music in the edit anyway.

Budgeting a test batch becomes straightforward once those rates sit in front of you. Ten UGC takes of 10 seconds on Seedance 2.0 Fast in 720p bill 1600 tokens, $16.00, before any premium model is touched. Promote the two best hooks to a 1080p master on Seedance 2.5 at 8 seconds each and you add 1200 tokens, $12.00. The whole experiment, twelve generations, closes at 2800 tokens, $28.00, as billed in SociaLover, pay as you go.

Cheaper per usable second
  • • Draft in 480p on Seedance 2.5 at 14 tk/s, then master the keeper once in 1080p
  • • Cut the brief to the seconds the hook needs: a 6 s Veo 3.1 spot bills 318 tokens, a 10 s one bills 530
  • • Test motion on Seedance 1.5 Pro at 3.5 tk/s or Veo 3.1 Lite at 7 tk/s before any premium pass
  • • Generate silent when a voiceover will replace the audio: Kling 2.6 drops from 19 to 9 tk/s
Where tokens leak
  • • Iterating in 1080p: every discarded Seedance 2.5 take bills 75 tk/s instead of 14
  • • Paying for native audio in a language it cannot speak: Kling only voices English or Chinese
  • • Running b-roll on a 53 tk/s model when a 13 tk/s tier covers the shot
  • • Stretching a 6 second idea to a longer clip: every extra second bills at the full rate

Frequently asked questions

How much does one second of AI video cost in 2026?
Between 3.5 and 75 tokens per second, which is $0.035 to $0.75, as billed in SociaLover, pay as you go, at 1 token = $0.01. The floor is Seedance 1.5 Pro in 720p without audio; the ceiling is Seedance 2.5 in 1080p. Most mainstream tiers, like Veo 3.1 Fast, Wan 2.7 or Kling 2.6, sit between 9 and 20 tokens per second.
Are these the providers' official API prices?
No. Every figure on this page is the price as billed in SociaLover, pay as you go, where 1 token equals $0.01 and the token amount covers the generation end to end. The providers publish their own API list prices, which differ and move on their own schedule, so this table should never be read as their raw API pricing.
Why is Sora 2 not in the table?
Because it no longer exists as a product. OpenAI announced the shutdown on March 24, 2026, closed the app on April 26, 2026, and cut the API on September 24, 2026. Before that, Sora 2 billed 13 tokens per second in 720p and Sora 2 Pro between 40 and 93. Any 2026 price comparison that still lists it is out of date.
Does 1080p really cost that much more than 720p?
On most models, yes. Seedance 2.5 moves from 31 to 75 tokens per second between 720p and 1080p, Seedance 2.0 from 20 to 49, Grok Imagine 1.5 from 19 to 33. Veo 3.1 Fast is the gentle exception, from 13 to 16. Flat rate models like Veo 3.1 or Kling 3.0 bill the same at both, so the rule is simple: draft low, master high.
How much does audio add to the price?
It depends on the model. Seedance 1.5 Pro doubles, from 3.5 to 7 tokens per second in 720p and from 8 to 15 in 1080p. Kling 2.6 jumps from 9 to 19. Veo 3.1, Kling 3.0 and Happy Horse 1.1 include native audio in their rate. Note that Kling's audio only speaks English or Chinese; every other language gets translated into English.
What does a 30 second video cost in a single generation?
Only Seedance 2.5 renders 30 seconds in one pass. At its per second rates, that is 420 tokens ($4.20) in 480p, 930 tokens ($9.30) in 720p and 2250 tokens ($22.50) in 1080p, as billed in SociaLover, pay as you go. It can also return the last frame of a clip, so longer cuts chain several generations that start exactly where the previous one ended.

Conclusion

A generated second costs between $0.035 and $0.75 in October 2026, as billed in SociaLover, pay as you go, and the table above is the full map between those two ends. The teams that pay the least per usable second are not the ones on the cheapest model: they are the ones that draft at 14 tokens per second and master at 75 exactly once.