2026-11-13 · VideoFreeTier

Dreamina Seedance 2.5 long video: the landing page says 180 seconds, the API says 30 seconds in one request and extends over multiple rounds

On the pages we read on 13 November 2026, Dreamina advertises a 180-second Seedance 2.5 clip through a beta long-video mode, while the BytePlus ModelArk API tutorial says one request generates up to 30 seconds and longer videos come from multiple extension rounds. No official page we read joins those two numbers, and no page prints what a 180-second clip costs in Dreamina credits. The API pricing page does publish a token formula that adds input video duration to output video duration, so a chain of extensions is not simply six times one 30-second price (our arithmetic, with the assumption printed beside it). These are observations about named pages on named channels, not a statement that the model cannot make a 180-second clip. We held no Dreamina account, ran no generation and reached no checkout, so the real credit deduction is unpublished on the pages we read.

Two official pages give two different maximums

Read on 13 November 2026, the consumer page dreamina.capcut.com/seedance/seedance-2-5 prints 180 seconds in two separate pieces of copy, not one sentence rendered twice. Its title still reads Official Seedance 2.5: 4K & 30s AI Video Generator with Audio, and one feature block says You can create cinematic videos up to 30 seconds in standard mode or extend them to 180 seconds with the beta long-video mode. A separate paragraph says Seedance 2.5 is suitable for commercial storytelling with long-form generation up to 180 seconds. The word beta appears once on the page; we found no start date, end date, region limit or statement about what happens to 180 seconds when the beta ends.

The developer channel says something narrower. The ModelArk tutorial docs.byteplus.com/docs/en/modelark/seedance-2-5.md opens with It can generate up to 30 seconds of video in one request, extend videos over multiple rounds, accept up to 50 omni reference assets per request, and perform precise timestamp-based editing. Its capabilities section says the 30-second ceiling supports one-time generation of coherent videos up to 30 seconds long and presenting a complete story without multi-segment stitching.

These are two different claims on two different channels. The consumer page does not say 180 seconds is six 30-second rounds, and the API document does not say the consumer app has a separate long-video pipeline. We do not choose between them. A statement that Seedance 2.5 cannot exceed 30 seconds, or that a 180-second clip is definitely six stitched segments, would be stronger than either page publishes; this is a channel-level gap, not a model-level capability verdict.

What the API actually calls a video extension

The tutorial prints a configuration list for an explicit extension task. Set omni_reference_task_type to extend; include at least one reference video whose role is reference_video in content; set ratio to adaptive; and put at least one phrase such as extend forward/backward, continue or continue the story in the prompt. When the task type is set explicitly, the API validates these constraints earlier and returns an error immediately if they are not met; with auto the same constraints are checked after task creation.

A common summary of this section attaches a duration rule to the extension task, so it is worth being exact. The duration parameter itself has Default value: -1 and Value range: [4, 30] or -1. For video extension the tutorial says:

In video extension tasks, Seedance 2.5 selects the video to extend based on the prompt intent, then automatically keeps the output aspect ratio consistent with the selected video (ratio defaults to and only supports adaptive). You can specify the output duration or let the model determine it automatically (duration supports [4, 30] or -1).

The hard -1 rule belongs to video editing, not extension:

For Seedance 2.5 video editing tasks, duration only supports -1; you cannot specify a duration. The input video to be edited must be [4, 30] seconds long. Otherwise, an error is returned.

That last absence matters below. Every extension round is a request carrying a reference video, but the tutorial prints no per-round input length for extension, so any cost for a chain of rounds has to state its own assumption and let the reader change it.

The billing formula re-reads the clip being extended

The ModelArk video-generation pricing page read 13 November 2026 states Token prices are USD per million tokens and gives Estimated token consumption: (Input video duration + output video duration) × output video width × output video height × output video frame rate / 1024, with Estimated video price: Token unit price × token consumption. For dreamina-seedance-2-5-260628 the printed rates are $10.70 for 480p and 720p without video input and $11.7 for 1080p; with video input they are $6.40 and $7.0. The 4K column reads Not supported in both tables.

So the seconds of the clip being extended enter the same token count as the seconds of new output. That is a fact about the published formula on the API channel; it is not a statement about how the Dreamina consumer app charges credits. The same page says Token consumption values are estimates. Actual consumption is determined by usage.completion_tokens in the API response. It also applies a minimum token consumption when the input includes video, so short inputs can be billed above the raw formula, and it charges only for successfully generated videos.

Reproducing the published sample prices

Before extrapolating anything, we checked our reading of the formula against the official sample prices. At 720p 16:9 and 24fps, 1280 × 720 × 24 / 1024 = 21,600 tokens per second; at 1080p the same calculation gives 48,600 (our arithmetic). One token rate applies to the whole request, with input and output seconds added first.

All four endpoints match, which is why we use one rate per request. Splitting the same 720p example into input seconds at the no-video rate and output seconds at the with-video rate gives $5.3028 (our arithmetic), which does not match the published $4.838; we do not use that split anywhere on this page.

The low end of each with-video range is a floor, not a two-second price. At 720p, two seconds of input plus five of output would be $0.9677 by the formula (our arithmetic), while the printed low is $1.244. Four seconds of input gives $1.2442. The page labels that end Lowest price: 2-4 seconds input and separately documents a minimum token consumption, so we treat $1.244 as the minimum and do not divide it by two.

What a six-round chain would cost, and why the ratio is not fixed

Suppose a 180-second clip is one 30-second first segment plus five extension rounds of 30 seconds each, all at 720p, and each round feeds in S seconds of the previous clip. Every figure in this section is our arithmetic from the published formula and rates, not a price anyone quotes; the official pages do not say how long each round's input is, or that 180 seconds must be built this way.

The first segment at the no-video rate is 30 × 21,600 × 10.70 / 1,000,000 = $6.9336. Each extension round at the with-video rate is (S + 30) × 21,600 × 6.40 / 1,000,000. For comparison only, one hypothetical 180-second request at the no-video rate would be 180 × 21,600 × 10.70 / 1,000,000 = $41.6016; no official page offers that option, so it is a baseline we constructed, not a quoted price.

The multiple moves with S. A fixed claim such as saying an extended 180-second clip costs one and a half times a linear one is not something the published formula establishes; it depends on an input length the docs do not print. These are USD estimates on the API channel. Dreamina's consumer credits are a different unit, and the reconstructed per-second credit rate on another page of this site is itself an arithmetic reconstruction, so we do not multiply the two.

What the plan cards print, and what they leave out

The Dreamina USD pricing page at dreamina.capcut.com/pricing/dreamina-seedance-price was rendered on 12 November 2026, not on 13 November; we did not re-render it today. In that capture, the visible benefit list of each of the four cards, Basic, Standard, Advanced and Ultra, prints Extend video length immediately above See all benefits. No card states how many extensions are allowed, what one extension costs in credits, or what a 180-second clip costs in total.

The same page's FAQ says The number of credits required may vary depending on the model, feature, and generation settings you use. The Seedance 2.5 landing page FAQ says Yes. Dreamina provides free daily credits to test out Seedance 2.5 before you commit to a subscription plan. Before you buy, experiment with advanced capabilities like long-form video generation, multimodal inputs and reference-based creation to see how they match your creative workflow. Neither line prints a number of daily free credits. We did not hold an account, did not run a generation and did not reach a checkout, so the credit deduction for a 180-second clip is unpublished on the pages we read; we do not fill it with a third-party figure.

The help centre dreamina.capcut.com/help, read on 13 November 2026, has no occurrence of long video, long-form, extend or 180 in its visible text. The 180 strings in the raw HTML of that page are SVG path coordinates, not prose. We did not count the language entries on that page, so we make no claim about how many locales it offers. Each of these is an absence on one channel on one date, not evidence that the model lacks the feature.

Two reader comments, not a measurement

Captured on 13 November 2026 under a third-party tutorial video titled How to Use Seedance 2.5 Long Video Mode in Dreamina (1-3 Min AI Clips!), two readers describe the problem in their own words. Neither is a vendor statement, and we did not reproduce either claim.

@FarahAbd-b9f wrote Hello there, I’m using the Higgsfiled at the moment, can’t do a 1 minute video. Frustrating. I have to create at least 1 minute videos for my page , do you think this AI will pp will be better for me !? 1 minute is good enough for me. @jon_west wrote Almost every dark video I generate (either 30 seconds or long form) it has a murky sort of muddy grey noise over it. You can even see it in these clips just barely. Hopefully that gets sorted out soon, but man 3 minute videos are wild. They can get silly and you really have to control the prompt and use references well, though.

Together they show a demand for one-minute clips and a report of quality problems in long-form output. Neither comment carries a credit number, a measured 180-second generation or a per-round cost, so they cannot settle anything in the arithmetic above. We quote them to show the gap the pages leave open.

What we did not establish

Unconfirmed on 13 November 2026: whether the consumer 180 seconds is six 30-second rounds or a separate pipeline; the beta's start date, end date, region limits and post-beta behaviour; how many extension rounds a plan allows; the credit cost per round and the total credit cost of a 180-second clip; the size of the free daily credit grant; how many seconds of input each extension round actually uses; and whether the API-side USD estimate maps to Dreamina credits at all. We held no account, ran no generation and reached no checkout. Every number above is either a line printed on a named official page or an arithmetic result marked as ours, and the absences are stated per channel and per date.

A note on units: the API price table is quoted in USD per million tokens, while the consumer plan cards print credits. The two are different pockets; the help centre's own instruction is to check the current offer in your account. We did not convert between them and do not present any credit figure for a 180-second clip.

Before you plan a long video

Treat the two maximums as two separate published claims: 180 seconds is what the consumer landing page advertises, and 30 seconds in one request is what the API tutorial documents. If you need a long clip, plan for more than one round, and read the credit deduction shown in your own account before submitting rather than a copied figure. On the API side, check usage.completion_tokens in the response instead of trusting the estimate, and remember that input video duration is part of the token count. If a page prints Extend video length as a benefit but no count or cost, that is a benefit line, not a price.

The dated observations behind the model pages live on the data page. For the arithmetic on a single five-second clip and its per-second rate, see Seedance 2.5 API pricing; for the credit reconstruction on the four plan cards, see Dreamina Seedance 2.0 price; for the resolution columns that say Not supported, see Seedance 2.5 4K. We do not restate those figures here.