Skip to content

Video

Seedance 1.5 Pro

Video

Available on Ofrai — chat, images, and video in one place with shared credits.

94 creditsModel details

Video studio

Preview

No clip yet

Your latest clip preview will show here after generation.

Open studio

Example prompts

Choose an example to load its prompt and recommended settings into the Seedance 1.5 Pro creator above.

How it works

Create with Seedance 1.5 Pro in three steps

  1. 01

    Choose your inputs

    Keep Seedance 1.5 Pro selected, then add a prompt and any supported reference assets.

  2. 02

    Set the output

    Choose the aspect ratio, resolution, and available Video options before generating.

  3. 03

    Generate and refine

    Review the result, adjust the prompt or settings, and generate another version from the same place.

Doubao Seedance 1.5 Pro is the next generation in the same line, and the documentation describes it as a deliberate trade rather than a superset. Reference-image mode is gone — the reference states plainly that it is not supported. In its place is native audio: an `audio` parameter documented as unique to 1.5 Pro, which, set to true, returns a video with an AI-generated soundtrack attached. Everything else is recognisably continuous with the generation before it, and two small shifts are easy to miss.

The duration floor moves up: the range is now 4 to 12 seconds rather than 2 to 12, so the two-second sting that the previous generation would render is out of reach. And the plain image list gains a positional fallback — one image becomes the first frame, two images become first and last, and more than two silently drops everything after the first pair. If you are moving an existing workflow here, the reference-image capability is what you lose and sound is what you gain.

What 1.5 Pro gave up, and what it bought

The interface description lists four capabilities and one absence, and the absence is the most informative line in it. Supported: text-to-video, first-frame and last-frame image control, audio generation, and asynchronous task management with a task ID. Not supported: reference-image mode. That is not an oversight in the summary — it is stated as a property of the model, and the audio parameter is stated in the same breath as the feature unique to 1.5 Pro. So the generation is not strictly a superset of its predecessor.

A project that leaned on a reference image to hold a character's appearance across a shot loses the tool that did that, and the documented remedy is the opposite direction from what you would guess: there is no replacement reference slot to migrate into. What you get in exchange is audio generated natively in the same request, which removes a downstream step that most pipelines otherwise pay for in both time and money. The decision is therefore structural.

If identity fidelity was the reason you chose this line's reference mode, staying on 1.0 Pro Quality is a legitimate choice and this model is the wrong migration. If you were going to add sound in a later pass anyway, the trade is favourable.

Native audio is a flag, and it changes what a render is

The audio parameter is optional and documented with a single meaningful value, true. Setting it means the returned video includes an AI-generated companion audio track. The documentation does not enumerate separate options for dialogue, ambience or music, and it does not document a way to supply your own audio file — so the correct reading is that this is one on-or-off decision, not a mixer. That has real consequences for how you brief a shot.

You cannot tell the model what the scene should sound like in a request parameter, which means the prompt continues to do the work of describing content, and the sound arrives as the model's own interpretation rather than as something you directed. For a project where the audio is ambience under narration you add yourself, this is close to ideal and removes an entire stage.

For a project where a specific line has to land at a specific moment, the timing is not yours to set on this endpoint, and a two-generation approach — silent render plus audio in post — remains the safer plan. Decide which of those two projects you are in before the first render, because adding audio later to a model that produced it natively means discarding the take you already approved.

The positional fallback in the plain image list

This is the least obvious rule in the reference and it will silently change your result if you miss it. The image list accepts one to two image URLs, documented as compatibility rules rather than strict roles. One image is treated as `first_frame`. Two images are treated as `first_frame` and `last_frame` respectively. More than two, and only the first two are used while the rest are ignored — not rejected, ignored.

Because a request with extra images succeeds, the failure mode is a render that quietly used different assets than you thought, which is the most expensive class of bug to chase after the fact. The documentation's own advice is to prefer `image_with_roles` and specify the first and last frames explicitly, and that advice should be followed without exception on anything you care about. The compatibility list exists so that a request copied from an older workflow still runs, not so that you can rely on positional inference.

Two more constraints come with the list: as on the earlier generation, only URL format is accepted and base64 is not, and `image_urls` and `image_with_roles` still cannot be sent together.

The duration floor moved, and that decides which jobs qualify

Duration is optional, the documented values are 4, 5 and 12 seconds, the supported range is 4 to 12, and the default remains 5. Compare that with the range documented for the previous generation in the same family, which reached down to two seconds, and the shift is unambiguous: the bottom third of that range is gone. Nothing about the top of the range changed.

Practically this converts a set of jobs from natively supported to unsupported-by-length — the two-second sting, the one-beat insert, the whip transition between two cuts. There is no two-second request to attempt on this endpoint. The workaround is the obvious one and it is also lossy: render the shortest legal clip and trim inside it.

You pay render time for frames you delete, you choose where the trim lands after the fact rather than before it, and a transition trimmed from the front of a four-second render is a slightly different motion than one that was generated to end there. For anything time-critical, this is the single fact to check before planning a migration, and it is the difference between 'this endpoint does it' and 'this endpoint does it with an edit'.

Resolution and aspect ratios, and what stayed still

Resolution accepts 480p, 720p and 1080p with 720p as the documented default — the same three-step ladder as before, and, notably, documented without the restriction that attached 1080p to reference-image use in the previous generation. Since reference-image mode is not supported here at all, the interaction has nothing to attach to, which is the other half of why the ceiling is simply the ceiling. Aspect ratio is also carried over unchanged: 16:9 as the default, plus 9:16, 1:1, 4:3, 3:4 and 21:9.

The unchanged aspect ladder is worth more than it sounds during a migration, because it means a ratio change is not part of the work — a shot framed at 21:9 on the older generation is still framed at 21:9 here, and a vertical deliverable does not need re-authoring for a new default.

The practical workflow advice is to treat resolution as the only knob you are expected to move between the two generations, keep it at the 720p default while a shot is still in draft, and only commit to 1080p once the take you would re-render is one you have actually approved.

Prompting a model that may return sound

The prompt is required on every request and is documented as a description of video content, with the same guidance to detail scene, action and style. The documented example is again a beach sunset with golden light on the water and waves along the shore, which tells you the expected shape: concrete, physical, and weighted toward what is moving rather than toward abstract mood.

On this model the prompt has a slightly wider job than on its predecessor, because it is now also the only channel through which you can influence what the clip is about, with no reference slot carrying context for you. That argues for putting the setting and the subject's business in the prompt in full, rather than writing a sparse description on the assumption that an attached image would supply the rest.

It also argues for treating the audio as an output you review rather than one you directed: since the parameter is a boolean and the reference documents no control over what the track contains, the sound should be judged on whether it fits the shot, and if it does not, the honest fix is a silent render rather than a long prompt trying to describe a sound the model will not be steered by that way.

FAQ

Does 1.5 Pro support reference images?
No. The reference states that reference-image mode is not supported on this model. It supports text-to-video and first-frame/last-frame image control only. If a reference image is essential to your workflow, 1.0 Pro Quality remains the entry that documents it — along with its own restriction that 1080p is unavailable when a reference image is used.
What happens if I pass more than two images?
The list accepts one or two URLs by documented compatibility rule: one image becomes the first frame, two become first and last, and more than two uses only the first two with the rest ignored rather than rejected. Because the request still succeeds, you get a render built from assets you did not intend to use. Prefer `image_with_roles` and name the frames explicitly.
Can I still request a two-second clip?
No. The documented range here is 4 to 12 seconds with 5 as the default, against a 2-to-12 range on the previous generation in the same family. Two-second stings and single-beat inserts are no longer natively available and have to be trimmed from the shortest legal render.
How does the audio option work?
It is an optional parameter documented as unique to 1.5 Pro. Set to true and the returned video includes an AI-generated companion audio track. The reference documents no sub-options for dialogue, ambience or music, and no way to supply your own audio, so treat it as one on-or-off decision rather than a mixer.

FAQ

Can I use Seedance 1.5 Pro directly on this page?
Yes. The creator above is the working Ofrai studio with Seedance 1.5 Pro selected by default.
Where are my finished results saved?
Signed-in generations are kept in your Ofrai library so you can download or revisit them.
Which prompts work best with Seedance 1.5 Pro?
Describe the subject, setting, visual style, lighting, and camera or motion clearly. The examples above provide practical starting points.
Can I edit the example prompts before generating?
Yes. Load any example, then change its wording, aspect ratio, references, or other supported settings in the creator.

Compare other Video models that can handle similar creative work to Seedance 1.5 Pro.

Commercial use depends on the selected model and model terms.