Seedance 2.5 API and Pricing, Explained Through a Real Pipeline
Seedance 2.5 generates clips at any exact length from 4 to 30 seconds, and it does not price like the per-second models everyone quotes. Here is how it meters, which settings move the charge, what an exact-duration clip is worth inside a real pipeline, and how to reach the model through a published API endpoint.
Seedance 2.5 is ByteDance's newest video model, and the reason people keep asking what it costs is that it does not bill the way the premium models bill. Google's Veo 3.1 publishes a flat per-second rate, so the arithmetic is a multiplication. Seedance 2.5 meters by video tokens, which means the charge moves with resolution and duration together rather than with a single posted number. That is a better deal for the work most people do with it, and it is harder to quote in a headline.
This post covers what the model actually offers, how the billing works, which two settings decide your bill, and how to call Seedance 2.5 as a published API endpoint without writing a ByteDance integration first.
What you get before you talk about price
Seedance 2.5's spec sheet is unusual in one specific way, and it matters to the cost conversation:
| Capability | Seedance 2.5 |
|---|---|
| Clip length | Any exact duration, 4 to 30 seconds |
| Aspect ratios | 16:9, 4:3, 1:1, 3:4, 9:16, 21:9 |
| Resolution | 480p or 720p |
| Native audio | Yes, generated with the clip |
| Keyframes | First frame and last frame images |
The headline is the length. Most video models generate in fixed steps of a few seconds. Veo 3.1 accepts 4, 6, or 8 seconds and nothing between them. Seedance 2.5 accepts any whole number of seconds from 4 to 30, which means a clip can be built to match a narration track instead of the narration being cut to match the clip. Full details are on the Seedance 2.5 model page.
The trade is resolution. Seedance 2.5 tops out at 720p. For vertical short-form watched on a phone that is not a compromise. For a website hero video or anything shown large on a desktop, it is a real limit, and the answer is to draft here and re-render the winning prompt on Veo 3.1, which goes to 4K.
How Seedance 2.5 bills
The model meters per video token rather than per second of output. As of August 2026 the catalog rate is $0.0000107 per video token, roughly $0.0107 per thousand tokens. Token count scales with how many pixels you asked for and how long you asked for them, so the two levers on your bill are exactly the two settings you were already choosing:
- Resolution. 480p costs materially less than 720p for the same clip, because it is fewer pixels per frame across the whole duration.
- Duration. A 30-second single take is not the price of a 6-second draft. Exact-second selection cuts both ways: you never pay for seconds you will trim off, and you should not request 30 when the narration is 23.
Compare that to Veo 3.1, which posts $0.40 per second with audio at standard resolution, so an 8-second clip with audio is $3.20 no matter what is in the shot. That flat rate is easy to quote and expensive to iterate against. Seedance 2.5 sits well below the premium per-second models at 720p, which is why our own faceless channel pipelines render on it: trying five prompt variants to find the right shot is a routine step rather than a budget decision.
What you actually pay on Treza
Seedance 2.5 through a Treza pipeline is metered from a prepaid credit balance. One credit equals $0.01, packs start at $5, credits never expire, and there is no subscription of any kind. A failed generation charges nothing.
Two things make the metering concrete rather than a mystery:
- Before the run, the platform estimates the media cost of the graph from the nodes in it, taking clip count and duration into account, and refuses to start a run your balance cannot cover. You see the estimate rather than discovering the charge later.
- After the run, the run record carries the real cost, so every generation is auditable against the balance it drew down.
For sizing a budget across the whole catalog, the number we publish on the pricing page is $1.06, the median charge for one video generation across all models. Seedance 2.5 work at 720p typically lands under that median, and premium 4K renders land above it. Buy a $5 pack, run the shot you actually need, and the meter tells you the truth faster than any pricing table will.
Why exact durations save more than they cost
The interesting saving with Seedance 2.5 is not the per-token rate. It is the stitching you stop doing.
A 24-second Short built from 8-second clips is three generations, three prompts that have to agree on lighting and subject, and two seams where continuity can break. The same Short on Seedance 2.5 is one generation at exactly 24 seconds. You paid for one render, not three, and you did not spend an afternoon fixing a cut where the character's jacket changed shade.
When a piece does need multiple shots, a Sequence node stitches up to 12 of them into one master with the narration laid underneath, so long single takes and multi-shot assembly are both available in the same graph. The AI shorts generator template wires that pattern up already.
Calling Seedance 2.5 as an API without building the ByteDance integration
Direct API access to a frontier video model is never one HTTP call. It is credentials, an asynchronous submit-and-poll job pattern, retry logic for transient failures, handling content moderation rejections, and storage for the output before your product can do anything with it. That is real work before you have written a line of your own logic.
On a Treza AI video pipeline, Seedance 2.5 is a node on a canvas. You set the duration, aspect ratio, resolution, and audio on the node, wire it next to a script step, a voiceover step, or a captioning step, and publish the whole chain as a single versioned endpoint. Your application calls a typed /invoke route with JSON in and JSON out, or points an existing OpenAI SDK at the OpenAI-compatible route if that is what your stack already speaks.
The part that pays off later is the swap. When the next model wins the leaderboard, you change the dropdown on the node and republish. The endpoint your product calls does not change, your client code does not change, and nobody ships a migration.
The same pipelines are reachable from agents. Our hosted MCP server exposes them as tools that Claude and other MCP clients can call directly, so "render this brief on Seedance 2.5 and send me the link when it is done" is something an agent can carry out rather than describe.
Where Seedance 2.0 still wins
Seedance 2.5 is not a strict upgrade over its predecessor. Seedance 2.0 keeps reference image support, which is what you need when a product or a character has to look the same across a dozen generations. It caps at 15 seconds and goes up to 4K.
The practical split in our own graphs: Seedance 2.5 for long single takes and daily short-form volume, Seedance 2.0 when a specific product or face has to stay consistent, Veo 3.1 when the frame has to survive a full-screen desktop viewing. All three are nodes in the same canvas, so a pipeline can use more than one and pay each model's rate only for the shots it renders.
Start with one clip
The fastest way to answer "what does Seedance 2.5 cost for my work" is to run your real prompt at your real duration and read the charge. Open the AI video generator, pick Seedance 2.5, request the exact number of seconds your script needs, and generate once. Then decide whether the shot belongs in a pipeline that reruns it on a schedule, and build that graph around the node you just proved.
Frequently Asked Questions
How much does Seedance 2.5 cost?
Seedance 2.5 meters by video tokens rather than by a flat per-second rate. The catalog rate is $0.0000107 per video token as of August 2026, and the total for a generation scales with resolution and duration. On Treza it is charged from a prepaid credit balance where one credit equals $0.01, the run is estimated before it starts, and a failed generation charges nothing.
How long can a Seedance 2.5 clip be?
Up to 30 seconds in a single generation, at any exact whole-second duration from 4 to 30. That is the longest single-clip range in the Treza catalog and roughly four times Veo 3.1's maximum of 8 seconds.
Is Seedance 2.5 cheaper than Veo 3.1?
At 720p, yes, meaningfully. Veo 3.1 posts $0.40 per second with audio at standard resolution, which makes an 8-second clip $3.20 before you have iterated on the prompt. Seedance 2.5 sits well below that, which is why it suits daily volume production. Veo 3.1 earns its rate when you need 4K or maximum photoreal quality.
Does Seedance 2.5 generate audio?
Yes. Native audio, including ambience and effects, is generated with the clip, so drafts arrive watchable rather than silent.
What resolution does Seedance 2.5 output?
480p or 720p. For phone-first vertical content that is enough. For hero assets shown large, draft on Seedance 2.5 and re-render the winning prompt on Veo 3.1, which supports up to 4K. On a pipeline node that swap is one dropdown.
Can I call Seedance 2.5 through an API without a ByteDance account?
Yes. On a Treza pipeline, Seedance 2.5 is a node you configure once and publish as a versioned HTTP endpoint, with a typed /invoke route and an OpenAI-compatible route. A hosted MCP server exposes the same pipelines as tools for Claude and other agents, and there is no subscription: generations are metered from prepaid credits.
Do I need a subscription to use Seedance 2.5?
No. Treza has no subscription. You buy prepaid credit packs starting at $5, credits never expire, and each generation is charged from that balance at the model's metered rate.
