Seedance 2.5 API Pricing: 5 Clips Tested, Real Costs

Seedance 2.5 is one of the most important video model launches of the year. Multi-shot prompts in a single call, synchronised native audio, clips up to 30 seconds: that combination is rare, and it is why the model landed on so many pipelines within a week of release. What nobody seems able to tell you is what it costs to run.

The rate card is quoted per million output tokens. That is a sensible unit for a language model and close to useless when you are trying to work out whether a 30-clip-a-day social calendar costs $40 or $400 a month. So I put together the breakdown I wanted to read: how Seedance 2.5 pricing actually works, and then five real generations to check it against reality.

Five clips, deliberately different settings, every one fired at the live API. For each one you get the exact parameters, the video itself, and what it billed down to five decimal places. Total spend for the run: $14.81. If you only want the summary, Seedance 2.5 pricing is per second of output and resolution is the lever that matters. All five clips landed within 0.7% of the published rate card, which means you can budget straight off the table below.

How Seedance 2.5 pricing actually works

Seedance 2.5 bills on the generated video, not on what you send in. Prompts, first-frame images and reference images are all free as inputs. What you pay for is output tokens, at $10.97 per million.

Tokens are an awkward unit for video, so Segmind publishes the same thing as a per-second rate, and that is the number worth internalising. For a given resolution and aspect ratio, the per-second cost is fixed. Total cost is that rate multiplied by your clip length, anywhere from 4 to 30 seconds. There is no setup fee, no per-request minimum, and no separate charge for audio.

Three things follow from that, and all three held up across the run:

• Duration is linear. A 10 second clip costs almost exactly twice a 5 second clip at the same settings, and my 30 second clip billed 5.96x the 5 second one. No volume break, no penalty for long renders.
• Resolution is the expensive decision. 480p to 720p is a 2.25x multiplier at identical length and ratio. It is the steepest lever in the parameter space by a wide margin.
• Audio is free. generate_audio: true adds native synchronised sound and does not appear anywhere in the rate matrix. I left it on for every test.

Here is the full rate table from the Seedance 2.5 pricing page, covering every aspect ratio the model accepts.

The five tests, and what each one cost

I picked five configurations to span the price range rather than to flatter it: the cheapest thing worth generating, a standard delivery clip, a long vertical, a cinematic wide carrying three shots, and the 30 second maximum. Every clip below is unedited output. Each billed figure comes off the API response for that exact request, not from my own estimate.

Test 1: the cheapest useful clip, 480p at 5 seconds

This is the draft-pass configuration. When I am iterating on a prompt I do not want to pay delivery prices to find out the composition is wrong, so 480p is where the first several attempts belong.

At just over half a dollar this is cheap enough to run five or six variations of a shot before committing. The quality is visibly a draft: the bubbles read as mush at 480p and the marble texture smears wherever the water moves. That is the correct trade. You are buying an answer to "does this shot work", not a deliverable.

The billed figure also tells us something about the rate card. I expected $0.5325 and was billed $0.53135, a variance of -0.22%. The published per-second rates are rounded presentations of the underlying token maths, so treat them as accurate to a fraction of a percent rather than exact. For budgeting, that is noise.

Test 2: the delivery baseline, 720p at 5 seconds

Same length, same aspect ratio, same audio setting, one thing different: 720p instead of 480p. This is the single most useful cost comparison in the run, because the rate depends only on resolution, duration and aspect ratio. It never depends on what you asked for, so a different subject changes nothing about the price.

Five seconds at 480p billed $0.53135. The same five seconds at 720p billed $1.19463, which is 2.25x for an identical amount of footage. That is the steepest single lever in Seedance 2.5 pricing, and it is why drafting at 480p pays for itself immediately: every iteration you resolve cheaply is 2.2x saved.

For a marketing agency this is the workhorse setting. A five second product beat at 720p with usable audio, for a little over a dollar, is what I would build a social ad pipeline around. Thirty a day runs roughly $35.84.

Test 3: vertical for social, 720p 9:16 at 10 seconds

Reels, Shorts and TikTok want 9:16, and they want longer than five seconds. This test doubles the duration and rotates the frame, which lets me check two claims at once: that duration is linear, and that aspect ratio barely matters.

Both claims hold. The 9:16 rate at 720p is identical to 16:9 at $0.2389 per second, so rotating the frame cost nothing, and this ten second clip came in at 1.99x the five second version. Linear, as advertised.

For anyone shipping to several platforms, that means aspect ratio never needs to be a cost conversation: across the whole matrix the spread at 720p is under one percent. Choose 9:16 because the platform wants it. Watch duration instead, because that is where the money actually goes.

Test 4: cinematic multi-shot, 720p 21:9 at 15 seconds

Seedance 2.5 handles multi-shot prompts, and this is where the model earns its keep for anything film adjacent. I wrote three explicit shots into one prompt and gave it 15 seconds, which is five seconds per beat.

At $3.58 this is fifteen seconds of graded anamorphic footage carrying three distinct shots, for under four dollars. For previsualisation that reframes the conversation: a director wanting six versions of a sequence before committing a crew is looking at about $21.

One practical note: give multi-shot prompts room. Three beats across 15 seconds gives each shot five seconds to breathe and the transitions land. The per-second price is identical either way, so there is no cost reason to rush a multi-shot prompt into six seconds.

Test 5: the 30 second ceiling, 720p at maximum length

30 seconds is the longest single call Seedance 2.5 accepts, which makes this clip the most expensive request the API will let you make at 720p. It is also the real test of the linearity claim: if per-second billing holds at six times the length of test 2, it holds everywhere in the range.

The rate card said $7.1670 and the invoice said $7.118433, a variance of -0.68%. Against test 2 that is 5.96x the cost for 6x the footage, so duration billing is linear for planning purposes and the drift at the long end is very slightly in your favour. That drift is the token maths showing through: billing is on the output tokens the clip actually produces, and the published per-second rate is a rounded stand-in for it. Across all five clips the spread between rate card and invoice ran from +0.01% to -0.68%, and it widened with length rather than randomly.

Two things are worth knowing before you plan around 30 second calls. The first is that latency does not scale with duration the way cost does: this clip came back in 145 seconds against 107 seconds for the five second version, so six times the footage cost 36% more wall clock. Long renders are better value in time than in money. The second is that a single 30 second generation gives you one continuous take rather than a cut sequence unless you write the shots in yourself, so for anything with structure you are still better off with several shorter calls stitched in an edit. The per-second price is identical either way, which keeps that choice creative rather than financial.

All five, side by side

All five are text-to-video calls with audio on and the seed fixed at 42. Expected is the published per-second rate multiplied by duration. Billed is what the account was actually charged.

$14.80535 billed against $14.8845 expected, every clip inside 0.7% of the rate card. You can plan a Seedance 2.5 budget straight off the table above and be right to within a rounding error, which is more than I can say for most model pricing pages.

One more practical note from the run. Requests that fail validation before generation starts are billed at exactly $0.00, so an experiment that errors out costs you latency and nothing else. Worth knowing when you are still working out your parameters.

What actually moves the bill

Ranked by how much leverage each one gives you, based on the numbers above:

1. Resolution, by a mile. 480p to 720p is a 2.25x multiplier at identical settings. Nothing else in the parameter space comes close.
2. Duration, linearly. Strictly per second across the full 4 to 30 second range, which I checked at both ends. A 30 second 720p 16:9 clip is $7.12, the practical ceiling for a single Seedance 2.5 call.
3. Aspect ratio, barely. Under a percent of spread at 720p. Ignore it as a cost factor and pick what your platform needs.
4. Audio, not at all. It is free. Leave generate_audio on unless you specifically want a silent plate.

The optimisation that follows from this is unglamorous and effective: lock your seed, iterate at 480p until the prompt is right, then render the final at 720p once. Because the seed is deterministic, the 480p version is a genuine preview of the 720p render rather than a different roll of the dice. That one habit is worth more than every other lever on this list combined.

Planning a monthly budget

Four volume scenarios, all computed from the 720p rates and assuming a working month of 22 days.

• Solo creator, 5 vertical clips a day at 720p 9:16 for 10 seconds: $2.39 per clip, about $11.95 a day, roughly $263 a month.
• Agency, 30 five-second 720p ads a day: about $35.84 a day, roughly $788 a month.
• Studio doing previs, 6 fifteen-second 21:9 sequences a day: $3.60 per clip, about $21.61 a day, roughly $475 a month.
• Long-form, 3 thirty-second 720p clips a day: $7.12 per clip, about $21.36 a day, roughly $470 a month.

Every figure there comes off rates I verified against real billing rather than a calculator. The 480p-first discipline cuts all of them by more than half during the iteration phase, which is the single easiest saving available on this model.

Estimating cost in code

Since the rate is fixed per resolution and aspect ratio, you can price a request before you send it. I keep a small helper like this in front of any batch job so a bad loop cannot quietly spend a fortune.

RATES = {
    "480p": {"16:9": 0.1065, "4:3": 0.1046, "1:1": 0.1018,
             "3:4": 0.1046, "9:16": 0.1065, "21:9": 0.1065},
    "720p": {"16:9": 0.2389, "4:3": 0.2404, "1:1": 0.2389,
             "3:4": 0.2404, "9:16": 0.2389, "21:9": 0.2401},
}


def estimate_cost(duration, resolution="720p", aspect="16:9"):
    # Cost in USD of one Seedance 2.5 call.
    return round(RATES[resolution][aspect] * duration, 4)


estimate_cost(5, "480p")             # 0.5325, billed 0.53135
estimate_cost(15, "720p", "21:9")    # 3.6015, billed 3.58155
estimate_cost(30, "720p")            # 7.167,  billed 7.11843

The comments are my actual invoices against each estimate, which is the only reason I trust the helper.

And the call itself, with the parameters from test 2:

import requests

resp = requests.post(
    "https://api.segmind.com/v1/seedance-2.5",
    headers={"x-api-key": "YOUR_API_KEY"},
    json={
        "prompt": "A matte black mechanical wristwatch on a slab of split slate ...",
        "duration": 5,
        "resolution": "720p",
        "aspect_ratio": "16:9",
        "generate_audio": True,
        "seed": 42,
    },
    timeout=600,
)

print(resp.headers["x-cost"])              # exact amount billed, e.g. 1.194633
print(resp.headers["x-generation-time"])   # seconds spent generating
with open("out.mp4", "wb") as f:
    f.write(resp.content)

Three things to plan around. The endpoint is synchronous and returns the MP4 in the response body, so set a generous timeout: my clips took between 60 and 145 seconds. Every response carries an x-cost header with the exact amount billed for that request, which is where every number in this post came from, so you can log real spend per call instead of estimating it. And fire sequentially rather than in parallel if you are working through a batch from one process. Full parameter reference is on the Seedance 2.5 model page.

Honest assessment

What I like about this pricing model: it is predictable to the point of being boring, which is exactly what you want from a cost model. One multiplication gives you the price of any call, my five invoices matched the published rates to within 0.7%, failed requests cost nothing, free audio removes a decision you would otherwise make on every request, and the x-cost header means you never have to guess what a job spent.

What I would flag. There is no volume discount, so a team generating a thousand clips a month pays the same per second as one generating ten. 720p is the ceiling, so this is a previsualisation and social tool rather than a finishing one, and upscaling is a separate cost on top. Latency runs from about a minute for a short 480p draft to 145 seconds for a 30 second 720p render, which is fine for a queue and awkward for anything interactive, and because the endpoint is synchronous a long call holds a connection open the whole time. The published per-second rates are also rounded, so expect your invoice to land a few tenths of a percent either side of your estimate.

FAQ

How much does Seedance 2.5 cost per video?

It depends on length and resolution. A 5 second 720p 16:9 clip billed me $1.19, and the same length at 480p billed $0.53. Cost is the per-second rate multiplied by duration, from 4 to 30 seconds.

How is Seedance 2.5 API pricing calculated?

Seedance 2.5 API pricing is token-based on generated output, billed at $10.97 per million output tokens. Segmind publishes it as a fixed per-second rate for each resolution and aspect ratio, so in practice you multiply the rate by your clip length.

Does enabling audio cost extra on Seedance 2.5?

No. Audio generation is free and does not appear in the rate matrix. Setting generate_audio: true adds synchronised native audio at no additional charge.

Do longer Seedance 2.5 clips cost proportionally more?

Exactly proportionally. I billed $1.19463 for 5 seconds and $7.11843 for 30 seconds at the same settings, which is 5.96x the cost for 6x the footage. There is no discount for long renders and no penalty either.

What is the maximum Seedance 2.5 clip length and cost?

30 seconds. At 720p 16:9 that billed $7.11843, the most a single call can cost. At 480p a 30 second clip works out to about $3.20.

Does aspect ratio change the price?

Barely. At 720p the spread across 16:9, 4:3, 1:1, 3:4, 9:16 and 21:9 is under one percent. Pick the ratio your platform needs and treat it as free.

The short version

Five clips, $14.81 of real spend, and a pricing model that turns out to be genuinely simple once you stop reading it in tokens. Per-second rate times duration, and my invoices matched that to within 0.7% at every length from 5 seconds to 30. Resolution is the expensive decision at 2.25x from 480p to 720p, duration is strictly linear, audio is free, aspect ratio is noise, and failed requests cost nothing.

Draft at 480p with a locked seed, deliver at 720p once, and read the x-cost header so you always know what a job spent. You can run the same tests yourself on the Seedance 2.5 model page, and the full rate table lives on the pricing page. Every number in this post came out of a real billing header rather than a calculator.