Seedance 2.0 Cost Per Second: The Real Billing Model

Seedance 2.0 bills by output duration. You pay for the seconds of finished video the model returns, not for how long a GPU ran, how long your prompt was, or how large your input files were. On reAPI the Seedance 2.0 cost per second runs from $0.045 to $1.04, and which end of that range you land on is decided by three choices you make in the request.

Two of those choices are obvious: speed tier and resolution. The third one is not, and it is where most of the money leaks. There is a cheaper rate that applies only when you upload a source video, and passing images or first/last frames does not qualify, even though it feels like it should.

TL;DR

• Billing is per second of output, with duration clamped to a 4 to 15 second range.
• Three dimensions set the rate: speed tier (standard or Fast), resolution (480p through 4K), and whether the request carries an uploaded source video.
• The uploaded-video rate is roughly 39% cheaper at 720p, $0.125/s against $0.205/s.
• It only triggers on video_urls. Image inputs and first/last frame references stay on the higher rate.
• A 5-second 720p clip costs $1.03 from a text prompt, or $0.63 if it is driven by an uploaded video.
• Fast at 480p is the cheapest path at $0.078/s, or $0.045/s with an uploaded video.

The full rate card

Rates below are per second of output video^[1].

Two structural facts fall out of that table.

Resolution is the steepest lever. Going from 480p to 4K on the standard tier multiplies the rate by about 11. Going from standard to Fast at the same resolution saves roughly 18 to 20%. If a clip does not need to be 4K, resolution is the first thing to cut, not the tier.

Fast stops at 720p. There is no Fast path to 1080p or 4K, so any high-resolution work is on the standard tier by definition.

The rate that catches people out

The cheaper column is labeled "& uploaded videos," and the label is precise in a way that is easy to misread.

That rate applies only when the request carries an uploaded source video. In API terms, only when video_urls is populated. Every other kind of reference material, an image, a first frame, a last frame, audio, leaves you on the base rate.

This matters because image-to-video and first/last-frame workflows feel like reference-driven generation. You are supplying source material, the model is transforming it rather than inventing from nothing, and the intuition is that it should cost less. It does not.

At 720p the gap is 39%: $0.205 per second from a prompt or an image, against $0.125 per second when a source video is attached. On a 10-second clip that is $2.05 against $1.25.

The practical consequence is a forecasting error rather than a bug. Teams model their costs on the cheaper column because their workflow "uses references," then the invoice arrives priced on the base column. If you are budgeting an image-to-video pipeline, budget it at the from-prompt rate.

Worked examples

Multiply the rate by output seconds. That is the whole model.

A 5-second 720p clip from a text prompt: 5 × $0.205 = $1.03

The same clip driven by an uploaded video: 5 × $0.125 = $0.63

A 10-second 1080p clip from a prompt: 10 × $0.510 = $5.10

A 5-second 480p draft on Fast: 5 × $0.078 = $0.39

A 15-second 4K hero shot: 15 × $1.040 = $15.60

At volume, the shape of the spend matters more than the rate. A thousand 5-second 720p clips a month from prompts runs about $1,025. The same thousand clips at 480p on Fast runs about $390. The same thousand at 1080p runs about $2,550.

The formula for a mixed workload:

monthly cost ≈ Σ (output seconds × rate for that resolution, tier, and input mode)

Duration is clamped to between 4 and 15 seconds, so a request for 3 seconds bills as 4, and there is no single-call path past 15.

Picking a configuration

Draft on Fast at 480p. At $0.078 per second, a 5-second iteration costs 39 cents. Iterate there, then render the approved shot once at the resolution you actually ship.

Only pay for 1080p and 4K on final output. The 480p-to-4K multiplier is roughly 11x. A 4K draft is the most expensive mistake available in this pricing model.

If your pipeline can be video-driven, make it video-driven. The 39% discount at 720p is real, but only for video_urls. If you are already producing a source clip, feeding it in rather than describing it saves meaningfully.

Do not assume image-to-video is discounted. Budget it at the from-prompt rate.

Calling it and knowing the bill

The endpoint is async: submit returns a task_id, then poll.

curl https://reapi.ai/api/v1/videos/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2.0",
    "prompt": "a slow dolly through a neon-lit alley after rain, cinematic",
    "resolution": "720p",
    "duration": 5
  }'

That request bills at the from-prompt rate. Adding video_urls moves it to the cheaper column:

curl https://reapi.ai/api/v1/videos/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "doubao-seedance-2.0",
    "prompt": "restyle this shot as neon-lit and rain-soaked, keep the camera move",
    "video_urls": ["https://example.com/source.mp4"],
    "resolution": "720p",
    "duration": 5
  }'

Two platform rules worth knowing. Media inputs are public http(s) URLs only, with no base64 accepted on any model. And rates move, so the live table on reapi.ai/models/seedance-2-0 is canonical rather than any figure quoted in an article, including this one. Request and response shapes are in the reapi.ai/docs/seedance-2-0 reference.

FAQ

How much does Seedance 2.0 cost per second?

From $0.045 to $1.04 per second of output, depending on speed tier, resolution, and whether the request carries an uploaded source video. A 720p clip from a text prompt is $0.205 per second^[1].

Does prompt length affect the price?

No. Billing is by output duration only. Prompt length, input file size, and server-side queue time do not change the charge.

Why is my image-to-video generation billed at the higher rate?

Because the cheaper rate applies only when an uploaded source video is present. Images, first frames, and last frames stay on the base rate even though they are reference material.

What does a 10-second 720p video cost?

$2.05 from a text prompt, or $1.25 when driven by an uploaded video.

Is Fast available at 1080p?

No. Fast covers 480p and 720p. Anything above that runs on the standard tier.

What is the minimum and maximum duration?

Duration is clamped between 4 and 15 seconds, so shorter requests bill as 4 seconds and there is no single-call path beyond 15.

How do I estimate a monthly bill?

Sum your expected output seconds per configuration and multiply each by its rate. A thousand 5-second 720p prompt-driven clips is roughly $1,025 a month.

Where is the authoritative price?

The live table on the Seedance 2.0 model page. Rates change, so treat any number in an article as a planning figure.

Forecasting the bill instead of discovering it

The Seedance 2.0 cost per second is simple arithmetic once you know which of the four columns you are actually in. The mistake that costs real money is not misreading the rate card, it is assuming that supplying reference material puts you in the cheaper column when only an uploaded video does that.

So the useful habit is small: before you model unit economics, check whether the pipeline sends video_urls or an image. Draft at 480p on Fast, reserve 1080p and 4K for shots that ship, and multiply expected output seconds by the live rate rather than the one you remember.

References

1. reAPI. Seedance 2.0 model page — live per-second rates by resolution, tier, and input mode. Retrieved July 2026 from reapi.ai/models/seedance-2-0

Further reading

• reAPI. Seedance 2.5 features. reapi.ai/blog/seedance-2-5-features
• reAPI. Seedance 2.0 endpoint reference. reapi.ai/docs/seedance-2-0
• reAPI. Model catalog. reapi.ai/models