If you are preparing to use Seedance 2.5, the most important step is not writing a long list of adjectives. First decide which task mode fits the job, whether your references meet the limits, and whether the output settings match the final use.
This guide is for creators and teams using the current Seedance 2.5 web workspace. It is an operating guide, not an independent benchmark, API reference, licensing guide, or commercial-rights statement. The capabilities, limits, and constraints below are based on the current Seedance 2.5 video workspace and its provider request contract. They may change with provider configuration or product updates. Claims about quality, speed, pricing, or commercial rights are not included unless they are confirmed in the workspace.
Quick answer: choose the job before the settings
Start by naming the job, not the aesthetic. Use reference when supplied media needs to constrain the shot; use edit to change an existing source clip; use extend to continue a source clip before or after its current boundary; and use first/last frames only when the beginning and ending images matter more than video or audio reference.
Then make the first run narrow: one subject, one primary action, one camera instruction, and a duration you can judge. The companion Seedance 2.5 Prompt Guide explains how to turn that brief into an iterative shot prompt without presenting editorial examples as model facts.
What Seedance 2.5 currently supports
The current workspace exposes four task types: auto, reference, edit, and extend. You can start with text and combine image, video, and audio references when the task calls for them.
These are the capability boundaries shown on the current page:
| Item | Current supported range |
|---|---|
| Duration per run | 4, 5, 6, 7, 8, 9, 10, 12, 15, 20, 25, or 30 seconds; automatic duration is also available |
| Automatic duration billing | The model selects the duration, but billing uses the 30-second ceiling |
| Resolution | 480p, 720p, and 1080p |
| Reference images | Up to 30 |
| Reference videos | Up to 10, with 30 seconds total |
| Reference audio | Up to 10, with 30 seconds total |
| Output container | MP4 or MOV |
| Task types | Auto, reference, edit, and extend |
| Image roles | Reference image, first frame, or last frame |
“Supported” means that the request contract accepts the option. It does not mean that every combination of inputs will produce the same result. Source quality, prompt clarity, action complexity, and shot design still affect whether the output matches your intent.
Choose the task type first
The same prompt solves different problems in different modes. Choosing the mode before submitting can prevent avoidable failed attempts.
| If the job is to... | Start with | Contract boundary to check | Do not choose it when... |
|---|---|---|---|
| Explore an idea without a fixed input plan | Auto | Other request limits still apply | You already need a source clip or a specific media role |
| Preserve a subject, product, scene, or rhythm from supplied media | Reference | Up to 30 images, 10 videos, and 10 audio clips; video and audio each total up to 30 seconds | You need to modify or continue a source clip |
| Change an existing clip | Edit | Requires automatic duration and adaptive aspect ratio | The goal is to add a new section before or after the clip |
| Continue a source clip before or after its current boundary | Extend | Requires at least one reference video, automatic duration, and adaptive aspect ratio | You only need a standalone text- or image-led shot |
| Specify a beginning and ending composition | First/last frame images | Requires adaptive aspect ratio and cannot be combined with video or audio references | Existing video or audio must guide the request |
Auto: let the workspace choose the path
Auto is useful when you do not yet have a fixed combination of inputs. Describe the subject, action, and camera, then add references as needed. Treat it as an exploration entry point, not as a way to bypass the other request limits.
Reference: use media as visual constraints
Reference mode fits work that already has a visual sample for a product, character, scene, or style. After uploading, refer to the files in the prompt with @image1, @video1, or @audio1; the number follows the upload order.
A reference image is more than decoration. It can clarify the subject's appearance, composition, clothing, material, or scene relationship. If an image needs to define where the shot starts or ends, assign it the first-frame or last-frame role, keeping the additional constraints below in mind.
Edit: change an existing video
Edit mode is for an existing video plus a specific instruction about what should change, continue, or be adjusted. The current contract has two important constraints for edit tasks: the output duration follows the source clip, so duration should be automatic; and the aspect ratio should be adaptive.
If the goal is to continue the story after a clip rather than modify the clip itself, use extend mode instead.
Extend: add content before or after a clip
Extend mode adds content before or after a source clip and requires at least one reference video. It also requires an adaptive aspect ratio and automatic duration so the system can handle the relationship to the source clip.
First-frame, last-frame, and reference conflicts
First and last frames are useful when you already know how a shot should begin and end. The current workspace defines two clear boundaries for these jobs:
- The aspect ratio must be adaptive when a first or last frame is used.
- First- or last-frame references cannot be combined with video or audio references.
Do not combine these two plans in one request:
- Use a first image and a last image to control the transition.
- Upload video or audio so the model can follow an existing time-based reference.
If you need both a source video and a still image as a constraint, split the work into two stages: use edit or extend for the time-based operation, then use an image reference in a separate request. This does not change the model's capabilities, but it keeps each request easier to diagnose.
A reusable prompt structure
A prompt does not need to read like advertising copy. For a controllable shot, describe the elements in this order:
- Subject: What person, object, or environment matters most in the frame?
- Action: What happens, where does it start, and where does it go?
- Camera: Is the camera locked, pushing in, pulling out, panning, tilting, or tracking? At what speed?
- Environment and light: What are the location, time, weather, key light, and material relationships?
- Pacing and duration intent: Is this a slow transition, a continuous action, or a shot that needs a pause at the end?
- Reference relationship: State whether
@image1is a subject reference or a first frame, and whether@video1is an action or scene reference.
For example, a product shot could be written like this:
A matte black insulated bottle from @image1 sits on a wet stone slab. The camera makes a slow low-angle push-in as droplets move down the bottle, with a softly blurred background. Keep the product centered for the first half, then shift slightly right during the final two seconds so a cool rim light appears along its edge. Keep the motion continuous and restrained; do not add text.
This example describes an intended task, not a promise about a particular output. For the first run, validate the subject, action, and camera before adding more style and detail constraints.
A practical workflow from prompt to clip
Step 1: define the final use
Decide whether the clip is for vertical social content, a horizontal product page, concept previs, or further editing. The use case informs the aspect ratio, output container, and whether an audio reference is useful.
Step 2: upload only references that solve a problem
The workspace allows many files, but “can upload” does not mean “should upload everything.” Choose references that answer a specific need:
- Character consistency: an image that clearly shows the subject's appearance and clothing;
- Product consistency: an image that shows shape, material, and important details;
- Action and rhythm: a short, clear video or audio clip;
- Shot transition: first and last frames, without adding video or audio references to the same request.
Also check file types and sizes. The current contract accepts common image formats including JPEG, PNG, and WebP, MP4/MOV video, and MP3/WAV audio. Per-file limits are 30 MB for images, 200 MB for video, and 15 MB for audio.
Step 3: choose duration, resolution, and aspect ratio
Do not default to the highest resolution. Start from the purpose of the clip and the complexity of the movement: a draft check can use a lower resolution, while final delivery may call for a higher one. The current Seedance 2.5 workspace lists 480p, 720p, and 1080p; it does not offer 4K.
Step 4: confirm the credit cost before submitting
This guide does not replace the live price display with a fixed number. The workspace configures credits by resolution and duration, shows the exact cost for the selected combination before submission, and recalculates it on the server. If a complete price grid is not configured, the combination is not submitted at an invented unknown price.
Step 5: change one variable per iteration
When a result misses the target, avoid changing the subject, camera, style, and duration all at once. A more diagnosable process is to change one variable per run:
- Subject drift: reduce conflicting references and state the subject and action more clearly;
- Wrong camera: move the camera instruction earlier in the prompt and remove secondary actions;
- Slow pacing: shorten the action chain and define the beginning, development, and ending;
- Wrong framing: change the aspect ratio first instead of trying to fix composition with more adjectives.
Seedance 2.5 vs. 2.0: compare only confirmed fields
The current landing page includes a workspace-contract comparison. It is not an independent benchmark and does not establish that every channel exposes identical model versions. When using another provider, follow that provider's documentation.
| Comparison | Seedance 2.5 | Seedance 2.0 listed on the current page |
|---|---|---|
| Maximum duration per run | 30 seconds | About 15 seconds |
| Reference images | Up to 30 | Up to 9 |
| Reference video | Up to 10, 30 seconds total | Up to 3, less than 15 seconds total |
| Reference audio | Up to 10, 30 seconds total | Up to 3, 15 seconds total |
| Audio-only reference | Supported | Not supported |
| Resolution | 480p / 720p / 1080p | 480p / 720p / 1080p / 4K |
| Output container | MP4 / MOV | MP4 |
The defensible conclusion from this table is narrow: within the current workspace contract, 2.5 is better suited to longer, reference-heavy jobs and to edit or extend workflows. The table does not prove that its visual quality, speed, or success rate is always higher in every scenario.
Common limits and troubleshooting checklist
When a submission fails, check these items first:
- Did you select 4K as though it were available for 2.5?
- Is the total duration of reference video over 30 seconds?
- Is the total duration of reference audio over 30 seconds?
- Is an edit or extend task using a non-adaptive aspect ratio?
- Does an edit task still have an explicit duration instead of automatic duration?
- Are first- or last-frame references present together with video or audio references?
- Is the selected duration and resolution missing from the configured price grid?
- Is a file over the per-file limit or in an unsupported format?
Frequently asked questions
How long can a Seedance 2.5 clip be?
The current workspace supports explicit durations up to 30 seconds, as well as automatic duration. Automatic mode is billed at the 30-second ceiling; actual availability remains subject to the current provider contract.
Can it generate 4K?
Do not treat 4K as an available Seedance 2.5 option in the current workspace. The page lists 480p, 720p, and 1080p for 2.5; 4K appears in the comparison fields for Seedance 2.0.
How many reference files can one task use?
Up to 30 images, 10 videos, and 10 audio clips. Reference video totals must stay within 30 seconds, and reference audio totals must also stay within 30 seconds. The selected combination must still satisfy the frame, task-type, and file-format rules.
How much does Seedance 2.5 cost?
This article does not publish a fixed number that may become stale. The workspace configures credit costs by resolution and duration and shows the exact cost before submission. Check the live display in the Seedance 2.5 video workspace.
Is this an official model announcement?
No. This is a usage guide based on the current website workspace and provider request contract. It does not replace the model provider's announcements, API documentation, content policy, or licensing terms. Check the relevant service terms and platform rules for commercial use, data handling, and copyright questions.
Final take: turn constraints into a workflow
The practical focus of Seedance 2.5 is not turning on every option. It is keeping the task type, references, camera intent, and output settings aligned. Start with a focused short task to validate the motion, then add references and duration gradually. That is easier to evaluate than submitting one complex request with conflicting goals.
For shot-writing templates and source-attributed workflow examples, continue with the Seedance 2.5 Prompt Guide. When your prompt or references are ready, open the Seedance 2.5 video workspace to check the current modes, parameters, and credit cost before submitting.
