Skip to main content

Seedance 2.5 Pro integration

Seedance 2.5 Pro is the latest Seedance video model, generating videos from text, images, reference media, or a multishot storyboard at 480p, 720p, and 1080p — with native audio, camera-motion control, and durations from 4 to 30 seconds.
Seedance 2.5 Pro is an AI video generation API that produces videos from a text prompt (text-to-video), from a first-frame image and optional last-frame image_end (image-to-video and transitions), from a set of reference_images, reference_videos and reference_audios (reference-to-video), or from a multishot storyboard of up to six shots. A single text prompt is always required. Compared with Seedance 2.0 Pro, the 2.5 model adds native audio generation, camera-motion presets, subject/style/video/audio references, multishot sequences, flexible durations from 4 to 30 seconds, and a selectable output container. Resolution is selected at generation time: there is a dedicated 480p, 720p, and 1080p generate endpoint. Task listing and status are per resolution — each resolution has its own list and status endpoint.

​
Key capabilities

  • Text-to-video and image-to-video: Generate from a text prompt alone, or guide generation with a first-frame image and optional last-frame image_end
  • First-to-last-frame transitions: Provide both image and image_end to generate a video that transitions between the two frames
  • Reference-to-video: Supply up to 30 reference_images (subjects, characters, products, styles), up to 10 reference_videos (motion, action or style to imitate), and up to 10 reference_audios, and cite them in the prompt with @Image1, @Video1, @Audio1, …
  • Multishot storyboards: Provide up to 6 multishot shots to drive a multi-scene sequence in a single request
  • Native audio: Generate sound effects and ambience with sound_effects, and optionally suppress background music with no_music
  • Camera motion control: Choose from more than 50 camera_motion presets, or keep a static tripod shot with camera_fixed
  • Resolution by generate endpoint: Dedicated 480p, 720p, and 1080p generate endpoints; resolution is not a body parameter
  • Flexible durations: Any integer from 4 to 30 seconds (default 5)
  • Seven aspect ratios: From ultra-wide film_horizontal_21_9 to ultra-tall film_vertical_9_21, default widescreen_16_9
  • Selectable output format: mp4 (default) or mov
  • Reproducible output: Provide a seed (0 to 4294967295) to repeat results; -1 selects a random seed
  • Safety filtering: Content safety checker is enabled by default via enable_safety_checker
  • Async processing: Webhook notifications or polling for task completion
  • 30-minute delivery: Generated video is returned via a URL valid for 30 minutes

​
Use cases

  • Premium marketing content: Produce high-resolution 1080p clips with native audio for ads, launches, and brand campaigns
  • Cinematic storytelling: Combine camera_motion presets, ultra-wide film_horizontal_21_9, and multishot sequences for film-style scenes
  • Character and product consistency: Keep the same subject or product across shots with reference_images, and carry over motion or style from existing footage with reference_videos
  • Longer-form clips: Generate up to 30 seconds in a single request
  • Social media content: Generate vertical social_story_9_16 clips for TikTok, Instagram Reels, and YouTube Shorts

​
Generation modes

​
API Operations

Generate a video using the 480p, 720p, or 1080p endpoint, then track it with the list and status endpoints for that resolution. Generation returns a task ID for async polling or webhook notification.

POST /v1/ai/video/seedance-2-5-pro-480p

Generate a 480p video from text, image, references, or a multishot storyboard

POST /v1/ai/video/seedance-2-5-pro-720p

Generate a 720p video from text, image, references, or a multishot storyboard

POST /v1/ai/video/seedance-2-5-pro-1080p

Generate a 1080p video from text, image, references, or a multishot storyboard

GET /v1/ai/video/seedance-2-5-pro-{res}

List tasks for a given resolution

GET /v1/ai/video/seedance-2-5-pro-{res}/{task-id}

Get task status and result by ID

​
Endpoint structure

​
Parameters

Resolution is determined by the generate endpoint you call (-480p, -720p, or -1080p) and is not a body parameter. The request body is identical for all three endpoints.

​
How reference videos are billed

Requests without reference_videos are billed on the output length alone — the requested duration, which is always at least 4 seconds. Requests with reference_videos are billed on the output length plus the combined duration of the reference videos, rounded to the nearest second and capped at 30 seconds — and that total is charged at the lower reference-to-video rate for the resolution. So a 4-second 720p video with one 10-second reference is billed as 14 seconds, not 4.
Reference videos change both the billed length and the per-second rate. If you are iterating on a prompt, trim the reference to the segment you actually need: the seconds you send are seconds you pay for. See the Pricing page for current rates.

​
Frequently Asked Questions

Seedance 2.5 Pro is the latest Seedance video model. You submit a text prompt (and optionally an image, reference_images, reference_videos, reference_audios, or a multishot storyboard) to a resolution-specific generate endpoint and receive a task ID immediately. Poll the GET status endpoint for that resolution or supply a webhook_url to be notified when the task completes, then download the video from the returned URL (valid for 30 minutes).
Resolution is selected by the generate endpoint, not by a body parameter. Call POST /v1/ai/video/seedance-2-5-pro-480p for 480p, POST /v1/ai/video/seedance-2-5-pro-720p for 720p, or POST /v1/ai/video/seedance-2-5-pro-1080p for 1080p.
Task listing and status are per resolution. Use GET /v1/ai/video/seedance-2-5-pro-{resolution} to list tasks and GET /v1/ai/video/seedance-2-5-pro-{resolution}/{task-id} to get a single task, using the same resolution you generated with.
Provide reference_images (up to 30), reference_videos (up to 10) and/or reference_audios (up to 10), then cite them in the prompt by index: @Image1 is the first item of reference_images, @Video1 the first of reference_videos, @Audio1 the first of reference_audios, and so on.reference_images and reference_videos are visual references and are mutually exclusive with the keyframe parameters image and image_end — send keyframes for a first/last-frame video, or references for a reference-to-video, not both. reference_audios is a separate modality and can be combined with keyframes.
No. The model composes its own soundtrack, guided by the rhythm and character of the audio you send: a 120 BPM reference produces a soundtrack at that tempo, but it is a new recording, not your file muxed into the result.So a reference with a clear beat, music or speech gives the model something to follow, while a flat tone or near-silence does not.
Each item of reference_videos is a publicly accessible HTTPS URL or a video upload file_id (upl_vid_...), and must be:
  • MP4 or MOV
  • 2 to 30 seconds long, with all references summing to 30 seconds or less
  • 24 to 60 fps
  • 200 MB or smaller
  • Between 409600 and 8295044 pixels (width × height)
  • An aspect ratio between 0.4 and 2.5
Videos are validated before the task is created, so a file that breaks any of these rules returns a 400 naming the field and is not charged.
multishot is an ordered list of up to 6 shots, each with its own prompt (up to 40000 characters) and optional duration (1 to 12 seconds). When provided, the shots drive the generation and the top-level prompt is still required. If any shot specifies a duration, every shot must, and the per-shot durations must sum to the top-level duration.
Yes. Native audio (sound effects and ambience) is generated by default (sound_effects: true). Set sound_effects: false for a silent video, or keep audio and set no_music: true to suppress background music while keeping other sound.
Use camera_fixed: true for a static tripod shot, or set camera_motion to one of 50+ presets (for example orbitLeft, pushIn, craneUp, whipPan, fpvDrone) to guide the shot. The preset is expanded into a descriptive instruction appended to the prompt.
Any integer from 4 to 30 seconds. The default is 5 seconds.
Seven aspect ratios: film_horizontal_21_9 (21:9), widescreen_16_9 (16:9, default), classic_4_3 (4:3), square_1_1 (1:1), traditional_3_4 (3:4), social_story_9_16 (9:16), and film_vertical_9_21 (9:21). This parameter is ignored when a first-frame image is provided.
Yes. Set output_format to mp4 (default) or mov. mov trades a larger file size for higher colour fidelity.
Yes. Provide a fixed seed value (from 0 to 4294967295) together with the same prompt and parameters to get consistent results. Use -1 (the default) for a random seed on each request.
Rate limits and pricing depend on your subscription tier and the selected resolution and duration. See the Rate Limits and Pricing pages for current values.Note that reference_videos change how a request is priced: the billed length becomes the output duration plus the combined reference duration (capped at 30 seconds), charged at the reference-to-video rate. See How reference videos are billed.

​
Best practices

  • Resolution choice: Use the -1080p endpoint for final, high-fidelity output and the -480p or -720p endpoints for faster, lower-cost iterations
  • Prompt writing: Be specific about the scene, subjects, motion, lighting, and visual style; when using references, cite them explicitly as @Image1, @Video1, @Audio1, …
  • References: Use reference_images to keep a character or product consistent, and reference_videos to carry over motion, action or style; both are mutually exclusive with the keyframes image and image_end
  • Reference video length: Trim references to the segment you need — their combined duration is added to the billed length (capped at 30 seconds)
  • Multishot: Either omit all per-shot durations (let the model distribute time) or set every one so they sum to the total duration
  • Audio: Native audio can occasionally be moderated; if a request is rejected for audio content, retry with sound_effects: false
  • Camera control: Set camera_fixed: true for stable product or talking-head shots, or pick a camera_motion preset for deliberate movement
  • Reproducibility: Fix the seed while iterating on a prompt so you can compare changes consistently
  • Production integration: Use webhooks instead of polling for scalable applications
  • Result retrieval: Download the output promptly, as the delivery URL is valid for 30 minutes
  • Seedance 2.0 Pro: Previous Seedance Pro generation with 480p, 720p, 1080p, and 4K generate endpoints
  • Seedance 2.0 Fast: Speed-optimized Seedance 2.0 variant
  • Kling 3: Higher-fidelity Kling 3.0 model for premium output