Overview

Seedance 2.5 is the new flagship Seedance model, providing high-quality video generation from text prompts, with optional image references to control subject appearance or style, plus video and audio references — all supported on the single seedance-2.5 task type (no VIP split).

Currently available models:

seedance-2.5

seedance-2.5-less-restriction

Pricing

Price per second(USD)

Model	480p	720p	1080p
seedance-2.5	0.15	0.35	0.80
seedance-2.5-less-restriction	0.165	0.385	0.88

With video references, billing is based on input video duration × (price per second / 2) + output duration × price per second.

Video editing tasks (omni_reference_task_type: edit, duration: -1) let the model choose the output length. Credits are reserved from your source video duration when the task is submitted, then settled against the duration of the video actually produced — if the result is shorter than the reservation, the difference is returned to your balance.

4K (2160p) is not available for seedance-2.5.

Note

Peak hours: during high-traffic periods, queue times may be extended.

Resolution: 480p, 720p or 1080p; defaults to 720p when omitted. 4K is not supported and will be rejected.

Duration: any integer from 4 to 30 seconds (default 5).

Reference images: the API supports using one or more images as references in your prompts. See Using Image References for syntax. Maximum 30 images.

Video/Audio references: maximum 10 videos, maximum 10 audios; total audio and video reference duration must not exceed 30s. Audio-only references are allowed.

Aspect ratio: 16:9 (default), 9:16, 4:3, 3:4, 21:9, 1:1, or adaptive. With adaptive the model selects the aspect ratio from the task type and the inputs. For video editing and video extension tasks adaptive is the only accepted value — see Video Editing and Video Extension below. For other tasks, the aspect ratio of the reference image takes precedence over the aspect ratio specified in the request parameters.

Video Editing and Video Extension

Besides reference-guided generation, seedance-2.5 can edit an existing video or extend one forward or backward. The model classifies the task from your prompt intent (for example "extend forward", "continue", or an instruction to change something in the source video), so these tasks have parameter requirements that differ from ordinary generation.

Declare the task with omni_reference_task_type to have those requirements validated when the task is submitted, instead of the task failing later.

`omni_reference_task_type`	Meaning	Requirements
`auto` (default)	The model infers the task from your prompt and the supplied references.	Parameters that do not match the task the model actually selects cause the task to fail after submission.
`reference`	Reference-guided generation from images, videos or audio.	No special `aspect_ratio` or `duration` requirement.
`edit`	Edits the visuals or audio of a source video.	At least one entry in `video_urls`; the source video must be 4–30 seconds; `aspect_ratio` must be `adaptive`; `duration` must be `-1`.
`extend`	Extends a source video forward or backward.	At least one entry in `video_urls`; `aspect_ratio` must be `adaptive`. `duration` is the requested output length (4–30 seconds).

Why `adaptive` is required

For editing and extension the model picks the source video to work from, and the output keeps that video's aspect ratio — so no fixed ratio can be honoured. adaptive is therefore the only accepted value for these two task types. For edit the output length also follows the source video, which is why duration must be -1.

If you declare edit or extend and omit aspect_ratio (and, for edit, duration), the required values are filled in for you. Supplying a conflicting fixed value is rejected with 400 rather than silently overridden.

Notes

The model re-checks the task type from the prompt while processing. If the task it identifies does not match the value you declared, the task fails — keep the prompt consistent with omni_reference_task_type.

For seedance-2.5-less-restriction editing requests, set auto_upload_assets to false so the source video URL stays directly readable; the source duration has to be measured to reserve credits.

Billing for editing follows the reservation-and-settlement rule described under Pricing above.

Example — extend a video forward

{
  "model": "seedance",
  "task_type": "seedance-2.5",
  "input": {
    "prompt": "Extend @video1 forward: the camera keeps drifting over the water as the sun sets",
    "mode": "omni_reference",
    "omni_reference_task_type": "extend",
    "video_urls": ["https://example.com/source.mp4"],
    "aspect_ratio": "adaptive",
    "duration": 8,
    "resolution": "720p"
  }
}

Example — edit a video

{
  "model": "seedance",
  "task_type": "seedance-2.5",
  "input": {
    "prompt": "In @video1, change the lighting to a warm golden hour",
    "mode": "omni_reference",
    "omni_reference_task_type": "edit",
    "video_urls": ["https://example.com/source.mp4"],
    "aspect_ratio": "adaptive",
    "duration": -1,
    "resolution": "720p"
  }
}

Using References in Prompts

When you include media in your request, you can reference them directly in the text prompt using @typeN placeholders, where N is the 1-based position within that media type.

Syntax

@image1, @image2, … – reference images

@video1, @video2, … – reference videos

@audio1, @audio2, … – reference audios

You can use multiple references in a single prompt to indicate different subjects or transitions.

The engine automatically converts @imageN / @videoN / @audioN to the upstream API format for compatibility. You may also use the Chinese format (@图片N / @视频N / @音频N) directly if preferred.

Examples

Single image reference

The cat in @image1 walks through a garden

Morphing effect

@image1 transforms into @image2

Multi-subject scene

The whale in @image1 meets the ninja in @image2

Video + Audio reference

The character in @image1 dances to @audio1 in the scene from @video1

Validation

If your prompt references media that was not provided (e.g., you use @image3 but only two images are supplied), the request will be rejected with a 400 Bad Request error.

Video References

seedance-2.5 supports video references via the video_urls array. You can provide up to 10 video URLs. The input videos are used as references for the generation.

How it works

Provide video URLs in the video_urls array (maximum 10; MP4/MOV only; maximum file size 200 MB per video).

Total video reference duration must not exceed 30 seconds.

Billing is based on (input video total duration + output duration) × price per second.

You may combine video references with image references (via image_urls) and audio references (via audio_urls).

Supported video formats: MP4 and MOV; each video must be 2–30 seconds and no larger than 200 MB. The input video must pass a content safety review.

Audio References

seedance-2.5 supports audio references via the audio_urls array. You can provide up to 10 audio URLs (MP3/WAV only; maximum file size 15 MB per audio).

How it works

Provide audio URLs in the audio_urls array (maximum 10; MP3/WAV only; maximum file size 15 MB per audio).

Total audio reference duration must not exceed 30 seconds.

Audio-only submissions are supported; audio references do not require an image or video reference.

Supported formats: mp3, wav.

Tips

Use publicly accessible media URLs (e.g., hosted on a CDN or cloud storage).

Keep reference media short (5–15 seconds) for best results.

Describe the desired output clearly in your prompt.

application/json

Bodyapplication/json

Examples

Task created (pending)Task completed

{
    "code": 200,
    "data": {
        "task_id": "07ed43de-465d-43b1-b62c-44249b14d818",
        "model": "seedance",
        "task_type": "seedance-2.5",
        "status": "pending",
        "config": {
            "service_mode": "",
            "webhook_config": {
                "endpoint": "",
                "secret": ""
            }
        },
        "input": {
            "prompt": "A woman sings and strums her guitar",
            "mode": "text_to_video",
            "duration": 5,
            "aspect_ratio": "16:9",
            "resolution": "720p"
        },
        "output": null,
        "meta": {
            "created_at": "2026-02-27T13:33:26.10346684Z",
            "started_at": "0001-01-01T00:00:00Z",
            "ended_at": "0001-01-01T00:00:00Z",
            "usage": {
                "type": "point",
                "frozen": 30000000,
                "consume": 0
            },
            "is_using_private_pool": false
        },
        "detail": null,
        "logs": [],
        "error": {
            "code": 0,
            "raw_message": "",
            "message": "",
            "detail": null
        }
    },
    "message": "success"
}

Seedance 2.5

Overview

Pricing

Note

Video Editing and Video Extension

Why `adaptive` is required

Notes

Example — extend a video forward

Example — edit a video

Using References in Prompts

Syntax

Examples

Single image reference

Morphing effect

Multi-subject scene

Video + Audio reference

Validation

Video References

How it works

Audio References

How it works

Tips

Request

Request Code Samples

Responses

Seedance 2.5

Overview#

Pricing#

Note#

Video Editing and Video Extension#

Why adaptive is required#

Notes#

Example — extend a video forward#

Example — edit a video#

Using References in Prompts#

Syntax#

Examples#

Single image reference#

Morphing effect#

Multi-subject scene#

Video + Audio reference#

Validation#

Video References#

How it works#

Audio References#

How it works#

Tips#

Request

Request Code Samples

Responses

Overview

Pricing

Note

Video Editing and Video Extension

Why `adaptive` is required

Notes

Example — extend a video forward

Example — edit a video

Using References in Prompts

Syntax

Examples

Single image reference

Morphing effect

Multi-subject scene

Video + Audio reference

Validation

Video References

How it works

Audio References

How it works

Tips