Seedance 2 pricing for text, image, and reference video

The selected mode determines which input roles are accepted. A first frame, reference image, video, or audio file adds a validated production input and may affect the provider request or quote. Where reference duration is used for pricing, CreateForge reads verified media metadata rather than accepting an untrusted client number.

Duration, resolution, native audio, returned frames, output format, and other controls can also be price relevant. Configure the actual shot before quoting. An article cannot safely provide one fixed number for every Seedance 2 video without ignoring those parameter differences.

The minimum shown in the facts panel represents the least expensive supported current configuration. It is a comparison aid, not a universal price. The protected quote displayed for the final request is the amount reserved when the user confirms generation.

Seedance 2, Seedance 2.5, Seedance 2 Fast, and Seedance 2 Mini are separate model entries. Confirm the selected variant before accepting a quote. Their names describe related products, not interchangeable pricing tiers under one task identifier.

Start with one continuous action and the shortest duration that can prove it. Use a review resolution where appropriate, then increase demands only after identity, camera, and motion are acceptable. A prompt containing several cuts and transformations may require more retries without producing a usable editorial sequence.

Record the protected quote, rejection reason, and prompt change for each attempt. Effective cost is the total credits used to produce an accepted shot. A focused input and review plan can reduce that total more reliably than choosing the lowest model minimum without testing the actual deliverable.

CreateForge reserves credits before it submits a Seedance task. A clear provider failure releases the reservation. An uncertain submission is not automatically sent again, because the external provider may already be processing it; reconciliation protects against duplicate video work and duplicate charges.

After success, eligible output is copied to controlled private storage and shown in the Library. Review that persisted asset for continuity and delivery. Provider completion, R2 storage, Library visibility, and ledger settlement are distinct states that the product reconciles into one user workflow.

Seedance pricing depends on the active parameters and may also depend on reference-video duration. Build the estimate only after deciding the generation mode, output duration, resolution, input roles, and whether a reference video is genuinely necessary. The server validates media duration and returns the protected quote for that exact normalized request. A rate copied from a table cannot replace that calculation when the source mix changes.

Budget references by their effect on acceptance, not by quantity. A clean first frame and a precise movement description may be more useful than several sources with competing identities, lenses, and motion. Assign every file one purpose and remove any source that does not constrain the result. This preparation can reduce diagnostic retries and also makes it easier to identify whether a failed clip came from the source, prompt, timing, or model behavior.

Track cost per accepted sequence after editing. Include generated clips that were not used, trimming, continuity repair, sound replacement, and the time needed to align neighboring shots. Compare Fast, Mini, 2.0, or another model with the same acceptance sheet rather than only the minimum quote. The right production choice is the workflow that reaches an approved sequence with predictable effort and a defensible total credit spend.

List every source, its role, verified duration where relevant, and the property it is expected to improve. Remove sources that are merely inspirational or duplicate another file. The quote can price the normalized request, but only the team can decide whether each input creates enough control to justify complexity and possible additional cost.

After the first accepted sequence, compare the planned role of every reference with its visible effect. If an input did not improve identity, motion, timing, or continuity, omit it from the next controlled test. This evidence-driven cleanup can simplify both creative diagnosis and budget forecasting without pretending that fewer inputs always produce better video.