Seedance 2.5 API Pricing: What a Clip Really Costs

A five-second Seedance 2.5 clip at 720p costs $0.67 on one API and $2.36 on another. Same model, same resolution, same length. Those two numbers sit side by side in Apiframe's provider comparison, last verified on 20 August 2026, and they explain why "what does Seedance 2.5 cost" has no single answer.

The arithmetic does have an answer. Seedance 2.5 is billed on output tokens that scale with pixels and seconds, each provider resells that meter at its own markup, and a few billing rules (resolution multipliers, video-input surcharges, minimum top-ups, expiring credits) decide what a finished clip costs. Below are dated, sourced numbers, turned into costs for single clips and real batches.

What Seedance 2.5 is and why its price moves around

Seedance 2.5 is ByteDance's video model. Volcengine announced it on 23 June 2026, pitching native 4K, 30-second output and 3D pre-visualization tools, and Caixin reported a price of over $28 for a 30-second 4K video, aimed at professional filmmaking and advertising. It follows Seedance 2.0, which launched in February 2026.

That $28 figure belongs to the 4K tier. At 480p and 720p, third-party rates sit between roughly ten and fifty cents a second. The model reached a wider developer audience when OpenRouter added it on 7 August 2026, with a headline rate of $0.1028 per generated second.

The specs that matter for pricing:

Clip length: 4 to 30 seconds per call, according to Segmind's test of five clips. RuntimeWire notes a clip can be extended twice.
Aspect ratios: 16:9, 4:3, 1:1, 3:4, 9:16, 21:9 and adaptive, per Atlas Cloud's launch post. Image-to-video is adaptive only and inherits the source image's ratio.
References: up to 30 images, 10 video clips and 10 audio clips, 50 assets in total. Seedance 2.0 allowed 9, 3 and 3, with 15 in total.
Audio: generated natively and free on the providers Segmind compared, which matters when you estimate a finished clip rather than a silent one.

How the meter works: tokens, pixels and seconds

Under every per-second price is a token count. Segmind publishes the formula:

tokens = (output height x output width x duration in seconds x 24) / 1024

At 720p in 16:9, a 1280 by 720 frame, that comes to 21,600 tokens per second of output. Multiply by the provider's price per million tokens and you have the per-second rate. Segmind charges $10.97 per million output tokens, so 21,600 tokens cost about $0.237 a second. fal charges $21.40 per million, which lands at about $0.462 a second for the same clip.

Three consequences fall straight out of the formula, and Segmind's billing confirmed each of them:

Cost is linear in duration. A 30-second clip costs six times a 5-second clip. No volume discount inside a single call.
Resolution is the big lever. Going from 480p to 720p multiplied cost by 2.25 in Segmind's tests. 1080p more than doubles it again: Segmind lists $0.5876 a second at 1080p against $0.2389 at 720p.
Aspect ratio barely matters. At 720p the difference between ratios was under 1%. A 9:16 vertical costs what a 16:9 costs, give or take a fraction of a cent.

Input tokens and audio generation are free on the providers Segmind compared. If you go to the first-party route, BytePlus ModelArk reports actual consumption in usage.completion_tokens in each API response, which is the number to log if you want to reconcile your own spend.

Per-second rates by provider, dated

These are list prices as published by each provider or by a comparison site. Most come from companies with a commercial interest in looking cheap, so treat them as a snapshot taken in August and September 2026 and recheck before you commit a budget.

Text-to-video and image-to-video, 720p

Atlas Cloud: $0.134 a second, flat across text-to-video, image-to-video and reference-to-video (Atlas Cloud).
Replicate: $0.2312 a second at 720p and $0.1028 at 480p, without video input (Atlas Cloud's comparison).
Segmind: $0.2389 a second at 720p 16:9 and $0.1065 at 480p (Segmind pricing page).
Apiframe: $0.29 a second. EvoLink: $0.293. WaveSpeed: $0.36 (Apiframe comparison, verified 2026-08-20).
fal.ai: about $0.473 a second at 720p and $0.2205 at 480p.
OpenRouter: from $0.1028 a second, with the provider table showing $0.2311 a second via Seed for both video-only and video-with-audio (OpenRouter, RuntimeWire).
Aitachyon: $0.20 a second at 480p and $0.44 a second at 720p at the time of writing, listed on the Seedance 2.5 model page.

On pure per-second price, Aitachyon sits near the top of this list, below fal and well above Segmind and Atlas Cloud. That is the honest reading of the table. What the extra buys is covered further down; if the lowest rate is your only criterion, the cheapest resellers above win.

1080p

Fewer providers publish 1080p rates. Segmind lists $0.5876 a second at 16:9, $0.5913 at 4:3 and 3:4, and $0.5904 at 21:9. Atlas Cloud's image-to-video page lists native 480p, 720p and 1080p, with 4K reached by upscaling through FlashVSR or ESR rather than natively.

Worked costs: 5, 10 and 30 second clips

Here is the same clip priced four ways. Segmind's figures are the amounts it was actually billed in its tests, which is better evidence than a rate card. The fal figures come from Segmind's side-by-side comparison, the Atlas Cloud figures from its own image-to-video page, and the Aitachyon figures are the per-second rate multiplied out.

5-second clip

480p: Segmind $0.53, fal $1.07, Aitachyon $1.00.
720p: Atlas Cloud $0.67, Segmind $1.19, fal $2.31, Aitachyon $2.20.

10-second clip

720p: Atlas Cloud $1.34, Segmind $2.38 (billed $2.37939 for a 9:16 clip), Aitachyon $4.40.
480p: Aitachyon $2.00. Segmind's rate works out to about $1.07.

30-second clip, the maximum per call

480p: Segmind $3.17, fal $6.45, Aitachyon $6.00.
720p: Atlas Cloud $4.02, Segmind $7.12, fal $13.87, Aitachyon $13.20.
1080p at Segmind's rate: about $17.63.

Segmind spent $14.81 in total on its five test clips. Single renders are cheap; volume and rerolls are where budgets go.

Batch math: what real production runs cost

Nobody renders one clip. The numbers that matter are for a batch, including the renders you throw away. Below, three batches at Aitachyon's rates at the time of writing, with the arithmetic shown so you can swap in any provider's per-second rate.

Batch 1: 20 hooks for a creative test

Twenty 5-second hooks at 720p: 20 x 5 s x $0.44 = $44.00. At Segmind's billed rate the same batch is about $23.89, at Atlas Cloud $13.40, at fal about $46.20.

If you are deciding what the 20 hooks should vary, the order in which to vary creative elements matters more than the render price.

Batch 2: a week of 30-second shorts

Seven 30-second clips at 720p: 7 x 30 s x $0.44 = $92.40. At Segmind that is about $49.84, at Atlas Cloud $28.14, at fal $97.09. If you add a separate narrator instead of relying on native audio, ElevenLabs voiceover on Aitachyon costs $0.043 to $0.086 per 450 characters at the time of writing, which is small next to the video line. The voiceover cost breakdown has the per-script math.

Batch 3: budget for rerolls

No sourced figure exists for how often a Seedance render needs a retake, so pick your own number and keep it explicit. As a planning assumption, say one clip in four gets rerolled because the motion or framing is wrong. Batch 1 then becomes 25 renders instead of 20: 25 x 5 s x $0.44 = $55.00. Batch 2 becomes roughly nine 30-second renders: 9 x 30 s x $0.44 = $118.80. Write the assumption into your budget so you can correct it after the first week.

Rerolls are different from failures. A render that errors out should cost nothing: Atlas Cloud says failed tasks are not charged, and Aitachyon refunds a failed render automatically, to the cent. A render that succeeds but looks wrong is billed everywhere.

A five-step process to cut the Seedance bill

This is the decision process worth pinning above your render script. Each step uses one of the pricing mechanics above.

Draft at 480p. 480p costs about 2.25x less than 720p. Render every variation at 480p first, since at this stage you are judging composition, motion and pacing, and texture can wait.
Promote only the winners to 720p. Re-render the two to five variations that survive review. On Batch 1 that means 20 hooks at 480p (20 x 5 s x $0.20 = $20.00) plus 4 winners at 720p (4 x 5 s x $0.44 = $8.80), for $28.80 instead of $44.00.
Keep drafts short. Cost is linear in seconds. Test a 30-second concept with its first 5 to 8 seconds before paying for the full length.
Do not pay for 1080p by default. On Segmind's grid 1080p costs about 2.5x the 720p rate. Reserve it for the final cut of a clip you know you will ship at that size.
Ignore aspect ratio when budgeting. The under-1% difference is noise. Render the ratio each placement needs and spend the attention on framing instead.

Before and after, same week of work

Before: 20 hooks at 720p, 5 seconds each, plus 7 full shorts at 720p, 30 seconds each. $44.00 + $92.40 = $136.40.

After: 20 hooks drafted at 480p ($20.00), 4 promoted to 720p ($8.80), 7 shorts tested with an 8-second 480p opener each (7 x 8 s x $0.20 = $11.20), then rendered in full at 720p ($92.40). Total $132.40. The saving on shorts is small because you still render them in full; the real saving is the shorts you kill after the 8-second test. Kill two of the seven and the total drops to $66.00 for the five survivors plus $40.00 of tests and hooks, $106.00 overall.

Video-to-video billing, the line that surprises people

When the input includes a reference video, the per-second rate falls and the billed seconds rise. Segmind's video-to-video rates at 16:9 are $0.0637 a second at 480p, $0.1429 at 720p and $0.3519 at 1080p. But billed seconds equal the output duration plus the greater of the total reference duration or two-thirds of the output duration, rounded up, per Segmind's pricing page.

Worked through for a 10-second 720p output:

With a 3-second reference: two-thirds of 10 is 6.67, rounded up to 7, which beats 3. Billed seconds = 10 + 7 = 17. Cost = 17 x $0.1429 = $2.43.
With a 12-second reference: 12 beats 7. Billed seconds = 10 + 12 = 22. Cost = 22 x $0.1429 = $3.14.
Plain text-to-video at the same length and resolution: $2.38.

So the lower rate is mostly offset by the extra seconds, and long references cost more than the clip they guide. Trim reference videos to what the model needs. On the first-party route, BytePlus ModelArk counts input duration in the token formula and applies minimum-token floors when the input includes video, so very short outputs with a video reference cost more than the formula alone suggests. Replicate also prices video input on separate, higher tiers.

Where to get Seedance 2.5 API access without a subscription

There are two routes, according to Atlas Cloud's access guide dated 25 August 2026: BytePlus ModelArk, the first-party international platform, or an aggregator that resells it.

First-party: BytePlus ModelArk

ModelArk publishes pricing docs, a Seedance 2.5 Series Pricing Calculator and a minimum-token table. It is the source every reseller marks up, so it is the reference point for cost. The trade-off is one more vendor account, one more billing relationship, and a model catalogue limited to one company's models.

Aggregators

Check these five terms before you top up, because they change the effective price as much as the rate does:

Minimum top-up. Atlas Cloud requires $25 minimum.
Credit expiry. Atlas Cloud credits expire 365 days after purchase. Money you load and do not spend in a year is lost.
Failure policy. Whether a failed render is billed, and whether the refund is automatic.
Failover. On OpenRouter only Seed serves Seedance 2.5, which RuntimeWire notes limits failover if that provider slows down.
Cost visibility. Whether each job shows the model it ran on and what it cost, so you can reconcile against the formula.

The subscription question mostly comes up with web studios that bundle credits into monthly plans. The comparison of credits and pay-per-call covers when a bundle is cheaper.

Running Seedance from code or from an agent

At volume, the web UI is the slow part. A batch of 20 hooks is a loop, and its cost should come back with the files:

A script or an agent receives the brief: 20 hooks, 5 seconds, 480p drafts.
It calls the video endpoint once per variation with the model, resolution and duration set explicitly, since those three parameters are the whole bill.
Each call returns a job with its cost. The script writes the file reference and the cost to a log, and sums the batch.
You review, pick winners, and the script re-renders those at 720p.

With an MCP server, step 1 becomes a sentence typed into Claude Code or Cursor and the agent handles the loop. On Aitachyon, connecting Claude Code is one command: claude mcp add --transport http aitachyon https://aitachyon.com/api/mcp --header "Authorization: Bearer ait_...". The same key works against the plain HTTP API. Every generated file gets a stable ref, such as scn_ for a scene, that you fetch with GET /api/generations/{ref}. Each API key also carries its own spend, an alert when it spends unusually fast, and a one-click revoke, which matters when an agent is the thing spending. The Claude Code MCP walkthrough goes through a full session, and the agent production setup covers chaining video, voice and images. The API and MCP quick start has the request format.

Seedance is also not always the right model for a shot. Kling v3 is listed at $0.16 a second silent and $0.32 with native audio, and Hailuo 02 at $0.085 a second, both at the time of writing. The Kling v3 vs Seedance 2.5 comparison covers which one to pick per job, and the model-by-model video API pricing guide lays out the rest.

FAQ

How much does Seedance 2.5 cost per second?

It depends on the provider and resolution. At 720p, published rates in August 2026 ranged from $0.134 a second at Atlas Cloud to about $0.473 at fal.ai, with Segmind at $0.2389 and Replicate at $0.2312. At 480p, rates start around $0.10 a second. Aitachyon lists $0.20 a second at 480p and $0.44 at 720p at the time of writing.

How much does a 30-second Seedance 2.5 video cost?

At 720p, between $4.02 (Atlas Cloud) and $13.87 (fal.ai) on published rates, with Segmind billed at $7.12 in testing and Aitachyon at $13.20. At 480p, Segmind's figure is $3.17. A 30-second 4K clip is a different tier, reported by Caixin at over $28.

Does Seedance 2.5 charge extra for audio?

Not on the providers Segmind compared: audio generation and input tokens were free on both Segmind and fal. OpenRouter's provider table also shows the same rate for video-only and video-with-audio via Seed.

Why is 720p so much more expensive than 480p?

Billing is based on tokens, and tokens scale with pixel count. A 720p frame has about 2.25 times the tokens of the 480p frame, and Segmind's bills showed exactly that multiplier. Aspect ratio changes the price by under 1% because it rearranges pixels without adding many.

Can I use the Seedance 2.5 API without a subscription?

Yes. BytePlus ModelArk and the aggregators listed here, including Atlas Cloud, Segmind, fal.ai, Replicate and OpenRouter, sell pay-as-you-go API access. Check minimum top-ups and credit expiry: Atlas Cloud, for example, requires a $25 minimum and expires credits after 365 days.

Sources

If you want Seedance next to Kling, Veo, Wan and the image and voice models on one prepaid balance that never expires, with every job itemised and failed renders refunded automatically, Aitachyon runs the same calls from the studio, the API or an agent, and the price of each call is published before you make it.