GPU cloud / Seedance API cost
Seedance 2.5 API cost calculator
Estimate pay-as-you-go output cost for Dreamina Seedance 2.5 on BytePlus ModelArk. This calculation assumes no input video, 16:9 output and 24 fps. The vendor bills actual output tokens, so the amount charged may differ.
| Resolution | Pixels | USD / 1M tokens | Official 5s example |
|---|---|---|---|
| 480p | 854 × 480 | $10.70 | $0.514 |
| 720p | 1280 × 720 | $10.70 | $1.156 |
| 1080p | 1920 × 1080 | $11.70 | $2.843 |
Source: BytePlus ModelArk pricing; Seedance 2.5 output specifications. Fetched 2026-09-27. The listed five-second figures are official examples, not guaranteed bills for every request.
Formula: estimated tokens = duration × width × height × 24 ÷ 1,024. Estimated USD = tokens × listed USD per million tokens ÷ 1,000,000. The result is rounded to three decimals. BytePlus says actual billing uses usage.completion_tokens.
How to limit wasted generation spend
Write a short shot brief first, including the intended aspect ratio, duration and essential action. Generate the shortest usable proof before committing to a longer final shot. The cost estimate rises with output duration and pixel count; compare settings for the same brief in the calculator above.
BytePlus documents a two-step Draft mode: the draft is billed as 480p, and the final task is billed again at its target resolution. The rate for the final step depends on whether the first step included input video; the draft itself is not counted as input-video duration. Review a draft before ordering a final render, and budget both charged tasks. This page's calculator estimates one normal task only; it does not add Draft mode costs or input-video minimum-token rules.
If you submit reference or input video, the applicable rate and minimum token rules change. Use the official pricing page and its Seedance pricing calculator for that workflow. Save the returned token count and invoice line for your own cost history. For self-hosted alternatives, compare the GPU inference calculator and GPU price data; those are different workloads and should not be treated as directly equivalent per clip.