Per-Token vs Per-Second AI Video Pricing: How to Calculate Seedance, Omni Flash and Veo Costs

TWno1 Editorial · Published 2026-11-01

Credits and specs follow TWno1's live billing settings. Last checked: 2026-10-05

As of October 2026, one second of 720p Seedance 2.5 video counts as about 21,600 tokens, or roughly US$0.231.

Google's Gemini Omni Flash counts the same second of 720p as 5,792 tokens. Veo 3.1 skips tokens entirely and charges by the second. All three call it a price. None of the units line up.

This guide lays out the three kinds of bill, gives you a formula you can paste into a spreadsheet, and prices one 8-second vertical shot across five models. Every rate is a list price from the vendor's own page, checked on 2026-10-05.

Three bills, three things being counted

Two of them state the same rule: failures are free. BytePlus does not charge tasks that fail or get blocked by moderation. Google bills only Veo videos that generate successfully. For Omni Flash, the pricing page we read (checked 2026-10-05) does not say.

Work out a Seedance bill yourself

BytePlus publishes the estimate:

tokens ≈ (input video seconds + output seconds) × width × height × fps ÷ 1024

Multiply by the rate per million tokens. Seedance 2.5 rates: US$10.70 at 480p/720p without video input, US$6.40 with it; at 1080p, US$11.7 and US$7.0.

Now check it against BytePlus's own 5-second 16:9 examples. Seedance 2.5 always runs at 24 fps.

All three rows match. The formula is safe to use. In a spreadsheet: =seconds*width*height*24/1024*rate/1000000.

Two things fall out of the math.

Aspect ratio does not change the price. At 720p, 9:16 is 720 × 1280 and 1:1 is 960 × 960. Both are 921,600 pixels, same as 16:9. A vertical short costs exactly what a widescreen one does.

The fps term is a constant. It looks like a dial. On Seedance 2.5 it isn't, because output is fixed at 24 fps. Only resolution and length move the bill.

Don't compare price per million tokens

Omni Flash charges US$17.50 per million output tokens. Seedance 2.5 at 720p charges US$10.70. Seedance looks about 40% cheaper.

It's the other way around. The two vendors turn one second of 720p into very different token counts, almost four times apart:

• Seedance 2.5: 21,600 tokens × US$10.70 ÷ 1,000,000 ≈ US$0.231 per second
• Omni Flash: 5,792 tokens × US$17.50 ÷ 1,000,000 ≈ US$0.101 per second (Google rounds it to about US$0.10)

A token is whatever each vendor says it is. It is not a shared currency. Google's own Gemini text-to-speech counts 25 audio tokens per second of speech, which has nothing to do with either video number. Always convert to dollars per second before comparing vendors.

One shot, five bills: 8 seconds, 9:16, 720p

This compares price only. Not picture quality, not features. Veo 3.1 Lite takes no reference images and cannot extend. Seedance 2.5 runs up to 30 seconds in one clip with up to 50 reference assets. The cheapest row may not be able to make your shot.

Move to 1080p and the order shifts:

Token pricing follows pixels. 1080p has 2.25 times the pixels of 720p, and the per-token rate is a little higher too, so the bill grows about 2.46 times. Veo's per-second pricing barely notices. Standard charges the same for both.

So: on Veo 3.1 Standard, if the shot is 8 seconds anyway, take 1080p. It costs nothing extra. On Seedance 2.5, lock the motion at 480p or 720p first, then render 1080p once.

Three ways the bill gets bent

The cheaper video-input rate on Seedance is a trap. With a reference video the rate drops to US$6.40 (US$7.0 at 1080p). But the input seconds count as tokens, and there is a minimum. BytePlus's example: a 5-second 1080p output with 2-30 s of input video costs US$3.062 to US$11.907. The top end checks out: (30 + 5) × 48,600 = 1,701,000 tokens × US$7.0/M = US$11.907. The bottom end does not. The formula gives (2 + 5) × 48,600 × US$7.0/M ≈ US$2.381, yet BytePlus lists US$3.062. That gap is the floor. For comparison, 5 seconds of 1080p with no video input costs US$2.843.

Per-second billing has a minimum unit too. A Veo 3.1 clip is 4, 6 or 8 seconds. 1080p, 4K and reference images all force 8. Need a 5-second shot? You pay for 6. Need it at 1080p? You pay for 8. Seedance 2.5 lets you pick any length from 4 to 30 seconds, which fits tighter.

Draft mode is two bills. Seedance 2.5's draft mode renders a 480p preview, then a 1080p final, and charges each. It saves you from paying for a bad 1080p render. It does not make 1080p cheaper. Draft once and decide: good deal. Draft every single shot by habit and the total goes up.

Which billing model wins when

• Short shots, low resolution, lots of retries. Seedance at 480p runs about US$0.103 per second and accepts any length from 4 to 30 seconds, so you never pay for padding.
• Fixed 8-second shots at 1080p. Veo 3.1's per-second price barely reacts to resolution. Fast adds 20%. Standard adds nothing.
• One long take. Seedance 2.5 goes to 30 seconds in one clip. Veo 3.1 needs 8 seconds plus extensions, and extension is 720p only. Remember that a 30-second take you reject still costs the full 30 seconds.
• Comparing vendors. Convert everything to dollars per second. Never compare per-million-token rates across vendors.

Limits of these calculations

• Omni Flash has only a 720p conversion. Google's price page gives 5,792 tokens per second for 720p. It does not give the figure for 1080p or other resolutions, and we don't guess it here.
• The formula is an estimate. BytePlus calls it one. It matched all three examples, but your console is the final word. The video-input floor is only shown for 1080p.
• Prices move. Every rate here is a 2026-10-05 list price. Time-limited discounts are left out.
• Price is not quality. We did not compare output. An expensive model is not automatically better, and a cheap one may not do the job.

The spreadsheet formula above has to be rebuilt for every vendor, and Omni Flash does not even publish its 1080p token rate. TWno1 puts the estimated points for a shot on screen before you submit it, so you skip the token-to-dollar step. If the generation fails, the points come back in full, the same way BytePlus and Google waive failed jobs; the rules are on the pricing page. If you need direct control over token parameters or heavy API volume, the vendor's own API is the better fit.

Related reading: Seedance Draft Mode: When a 480p Preview Before 1080p Actually Saves Money; Why Your AI Video Camera Won't Stay Still, and How to Rewrite the Prompt; Same AI Video Model, Different Price: What Resellers Charge for Veo 3.1 and Seedance 2.5.

FAQ

How do you calculate Seedance 2.5 tokens?

BytePlus publishes the estimate tokens ≈ (input video seconds + output seconds) × width × height × fps ÷ 1024. Seedance 2.5 always outputs 24 fps, so one second of 720p (1280 × 720) is about 21,600 tokens. At the 2026-10-05 rate of US$10.70 per million tokens, that is roughly US$0.231 per second.

How many tokens is one second of Gemini Omni Flash video?

Google's Gemini API price page (checked 2026-10-05) counts 720p Omni Flash output at 5,792 tokens per second, priced at US$17.50 per million output tokens. Google puts that at about US$0.10 per second. Input tokens are billed separately at US$1.50 per million.

Is per-token or per-second AI video pricing cheaper?

Neither by default. Using 2026-10-05 list prices for one 8-second, 9:16, 720p shot: Veo 3.1 Fast (per second) costs US$0.80, Omni Flash (per token) about US$0.81, and Seedance 2.5 (per token) about US$1.85. At 1080p, Seedance 2.5 costs about 2.46 times its 720p price while Veo 3.1 Fast costs only 20% more.

Does vertical 9:16 video cost more than 16:9?

Not under pixel-based token pricing like Seedance 2.5. At 720p, 9:16 (720 × 1280), 16:9 (1280 × 720) and 1:1 (960 × 960) all contain 921,600 pixels, so the token count and cost are identical. Veo 3.1 bills per second, so aspect ratio does not change its price either.

Why does Seedance cost more with video input when the rate is lower?

Input video seconds count toward tokens, and a minimum applies. In BytePlus's example, a 5-second 1080p output with 2-30 seconds of input video costs US$3.062 to US$11.907, while the same 5 seconds without video input costs US$2.843, even though the rate falls from US$11.7 to US$7.0 per million tokens.

References

• BytePlus ModelArk model pricing (checked 2026-10-05)
• BytePlus ModelArk: Seedance 2.5 (checked 2026-10-05)
• BytePlus: create a video generation task (API) (checked 2026-10-05)
• Gemini API pricing (checked 2026-10-05)
• Gemini API: Veo video generation (checked 2026-10-05)
• Google Cloud Text-to-Speech pricing (checked 2026-10-05)
• Runway API pricing (checked 2026-10-05)
• Kling API pricing (checked 2026-10-05)

• Pricing and credits
• Tutorials