Video route guide · checked Nov 13, 2026

Seedance 2.0 API pricing by provider and unit

Seedance 2.0 is sold through aggregators that meter it differently: some bill per generated second, some per thousand tokens, and some per request. The unit, not the headline number, decides which quote is cheaper.

Short answer: APIDir's accepted observations list Seedance 2.0 at $0.3034 per generated second on fal (720p with audio) and $0.205 per second on Kie.ai, both checked in August 2026. Kie.ai's pricing page retrieved on November 13, 2026 lists bytedance/seedance-2 from $0.057 per second and a fast variant from $0.034, while fal lists a reference-to-video product at $0.014 per 1,000 tokens. Convert every quote to cost per accepted clip before comparing.

Seedance 2.0 meters across providers. APIDir rows carry an August 2026 check date; provider rows were retrieved November 13, 2026.
RouteRateUnitSource date
seedance-2 on falvideo second, 720p + audio$0.3034per secondAPIDir accepted observation, 2026-08-22.
seedance-2 on kie-aivideo second$0.205per secondAPIDir accepted observation, 2026-08-21.
bytedance/seedance-2 on kie-aivideo second$0.057per secondKie.ai pricing page, retrieved 2026-11-13.
seedance-2 fast on kie-aivideo second$0.034per secondKie.ai pricing page, retrieved 2026-11-13.
seedance-2.5 on kie-aivideo second$0.085per secondA newer version at a different rate.
seedance 2 reference-to-video on fal1,000 tokens$0.014per 1k tokensfal pricing page, retrieved 2026-11-13.
seedance-2 on wavespeedrequest, fast 720p$1.00per requestAPIDir accepted observation, 2026-08-22.

Three units, one model name

A generated video contains seconds, pixels, frames, and audio, and different aggregators choose different ones as the billable unit. fal historically meters Seedance-style video per generated second at a named resolution and audio setting. Kie.ai publishes a per-second rate converted from its credit system. fal also sells a reference-to-video product metered per thousand tokens, which is a completely different economic object even though the model family name matches.

The practical rule is to convert each quote to cost per accepted clip at the resolution and duration you need. A per-second quote multiplies by clip length. A per-token quote requires knowing how many tokens a clip consumes at that resolution and duration, which usually means reading the model card or running one measured job. A per-request quote is a fixed cost per call but may hide a duration cap.

The observed spread is more than 5x

Take an eight-second clip. At APIDir's accepted fal rate of $0.3034 per second the clip costs $2.43. At the accepted Kie.ai rate of $0.205 the same clip costs $1.64. At the Kie.ai page rate retrieved in November of $0.057 the clip costs $0.46, and the fast variant lands at $0.27. Across those four quotes the same nominal eight-second output ranges from $0.27 to $2.43.

The spread does not prove one provider is overpricing. Resolution, audio, fast versus standard quality, queue priority, and the exact model version can all differ. It does prove that a video budget built on a single provider's headline number is not transferable. Record the full configuration beside each rate: provider, model version, mode, resolution, audio flag, duration, and whether the price is per second or per request.

Accepted-output math beats per-clip math

Video generation fails in ways that cost money: motion artifacts, prompt drift, unwanted text, and cuts that do not survive review. A workflow that accepts 70% of generations needs about 1.43 billable attempts per accepted clip. At $0.46 per attempt that is $0.66 per accepted clip; at $2.43 per attempt it is $3.47. The acceptance rate multiplies the provider gap rather than narrowing it.

Measure the acceptance rate on your own prompt set and rubric. Include the review time, because a cheap route that produces unusable motion is not cheap. Then compare cost per accepted clip across routes, not cost per generation. That single change usually reorders a shortlist.

Retries, seeds, and duration caps

Duration caps are a hidden unit. A provider that bills per second but caps clips at five seconds needs three calls for a fifteen-second output, and each call carries its own prompt and inference cost. Check the maximum duration before annualising a budget. The same applies to image-to-video and reference-to-video modes, which may accept different inputs and produce different outputs at different rates.

Retry policy matters too. If the provider does not charge for failed generations, a conservative retry loop is cheap. If every attempt is billable, a retry loop on an unstable configuration becomes the dominant cost. Read the provider's failure-billing terms and confirm them with one small invoice; do not infer them from a marketing page.

Worked monthly estimate: 100 clips

One hundred accepted eight-second clips at an 85% acceptance rate requires about 118 billable attempts. At the November Kie.ai standard rate of $0.057 per second that is 944 seconds, or $53.81. At the accepted fal rate of $0.3034 per second it is $286.41. At the accepted Kie.ai rate of $0.205 it is $193.52. The route choice moves a small campaign by an order of magnitude, which is why a recheck before each batch is worth the ten minutes.

The calculator below takes cost per billable attempt, an acceptance rate, and an optional fixed platform cost. Enter the per-clip price you derived from the provider's unit, not the per-second price, so the estimate stays unit-correct.

Recheck before you commit

The August accepted rows and the November page rates describe different points in time, and this guide does not assume the movement is permanent. Model catalogs reprice, rename variants, and change resolution defaults. Retrieve the current page, capture the exact configuration, and reconcile one invoice before a large batch. When a rate moves by a factor of three, the honest conclusion is that the earlier row is stale, not that the provider changed its business model.

For procurement, ask each provider for the same three facts: the billable unit, the configuration that unit assumes, and whether failed generations are charged. A provider that cannot answer all three in writing is not comparable on price alone.

Estimate cost per accepted clip

Enter the per-attempt price for one clip at your configuration, the acceptance rate you measured, and any fixed monthly platform cost.

Estimate cost per accepted clip

Enter the per-attempt price for one clip at your configuration, the acceptance rate you measured, and any fixed monthly platform cost.

Clips that pass your review and ship.
One clip at your duration and resolution, converted to USD from the provider's unit.
Accepted clips divided by all billable attempts.
Storage, delivery, or review tooling you already know.
Reset example

Planning result

Review the example inputs, replace each rate with the exact current value from the linked source, then calculate. APIDir does not send these values to a provider.

Questions developers ask

How much does Seedance 2.0 cost per second?

Quotes range widely by provider and configuration: APIDir accepted $0.3034 per second on fal and $0.205 on Kie.ai in August 2026, while Kie.ai's page listed $0.057 per second in November 2026. Verify the resolution, audio setting, and model version with each quote.

Why is one provider metering Seedance per token?

Aggregators choose the unit that matches their infrastructure. A per-token reference-to-video product is not directly comparable to a per-second text-to-video product until you convert both to cost per accepted clip.

Do failed video generations cost money?

It depends on the provider. Confirm the failure-billing terms and check one invoice before relying on a retry-heavy pipeline.

How many attempts should I budget per accepted clip?

Use your own measured acceptance rate. An 85% rate needs about 1.18 attempts per accepted clip; a 70% rate needs about 1.43.

Sources and check dates

Every rate in this guide was retrieved from the linked page on Nov 13, 2026. Prices can change; recheck the source before purchasing.