Seedance A Series

This page focuses on 2.0-A and 2.5-A. Both support image, reference-video, and reference-audio inputs; reference-video mode requires at least one video and allows optional audio.

Models covered

seedance2.0-A
seedance2.5-A

Capability matrix

Model	Output seconds	Resolution	Max images	Max videos	Max audio	Total references
`seedance2.5-A`	4–30	480p / 720p	30	10	10	50
`seedance2.0-A`	4–15	480p / 720p / 1080p	9	3	3	12

Modes and combinations

A2.0, A2.5 and E720 show “Reference video/audio”: at least one video, with optional audio. Fixed-30-second B shows “Reference video” and excludes audio. B2.0, fastB, F and C show “Reference audio”: at least one audio item, with no video or image mixed into that mode.

Ordinary reference-image mode needs at least one image; these audio-capable Seedance presets also explicitly allow attached audio. Text-to-video carries no references. First-frame needs one image; a frame pair needs exactly two distinct images. A2.0 requires paired frame materials and auto ratio. E/F, A2.5, fixed-30-second B and C currently expose no frame-pair mode.

Sources, files and reference duration

Series	Sources and per-file limits	Reference duration
A2.0	Public HTTPS URL; 100 MB per file	Video 1.8–15s each, 15s total; audio 2–15s each, 15s total; video pixel count 409600–8295044
A2.5	Public HTTPS URL; 100 MB per file	Video and audio each have a 30s total
B2.0 / fastB	HTTPS or complete Data URL; images 20 MB, audio 300 MB	Audio 15s total; video unsupported
E720 / F	HTTPS or complete Data URL; images 20 MB, video/audio 50 MB	Current preset: video (E720 only) and audio up to 15s each, 15s total per kind
C	Public HTTP/HTTPS URL; file-size limit unpublished	Video unsupported; reference audio is limited to 15s per file and in total
Fixed-30-second B	Public HTTPS URL; unique source per reference	No published reference-duration ceiling to reuse

A “Public” label means an HTTP(S) address is retained for submission, not guaranteed access without authentication or expiry. A-series further requires HTTPS. Ordinary clipboard content, crops and local files do not automatically obtain public URLs; this system adds no public upload service. Local encoded media works only for models explicitly accepting Data URLs. C rejects data: input with unsupported protocol: data.

Fields, defaults and differences

Field	Type	Meaning
`model`	string	Use a public ID from the table, never a retired model name
`prompt`	string	Describe movement and camera; use the reference numbering below
`seconds`	integer	Integer; duration is a compatible alias, but both must agree; C requires an explicit value and fixed-30-second B requires 30
`resolution`	string	Must belong to the model list; do not infer resolution or a quality field from its name
`aspect_ratio`	string	Usually 16:9, 4:3, 1:1, 3:4, 9:16, 21:9; A2.0 supports auto, B adaptive; E/F excludes adaptive; C currently offers 16:9/9:16
`references`	object[]	Object array; type=image/video/audio, role describes purpose, source is a URL or permitted Data URL
`task_mode`	string	B2.0/fastB only: text, first_frame, first_last_frame, references; E/F: text, references; do not copy Canvas mode values into this field
`generate_audio`	boolean	Output track only. C exposes it with a service default of false; it is independent of input audio. B service docs list a toggle but PopGo's current Canvas preset keeps it closed.
`return_last_frame` / `web_search`	boolean	Optional booleans for B2.0/fastB; not exposed in Canvas; unsupported by E/F
`priority`	integer	Integer 0–9 for B2.0/fastB; not exposed in Canvas or universal to other models

Current UI defaults: A2.0/A2.5 4s/480p; B2.0/fastB 5s/720p; E/F 5s at the model's fixed resolution; C 10s/720p; fixed-30-second B 30s/720p. Default ratio is 16:9. Send explicit parameters to avoid differences in service defaults.

Capability differences: A2.0 service material lists 4K, but PopGo retains the verified 480/720/1080p preset and keeps 4K closed. The public F1080 name differs from service material; retain F1080 with video count zero instead of borrowing another model's video capability. C follows the confirmed image/audio contract.

Unified references and examples

All models use references. Ordinary image/video/audio uses role: reference; C audio uses role: reference_audio, preserved in the outgoing request. Do not send a unified array together with multiple old aliases. Sources must be unique. Frame pairs use image items with first_frame/last_frame and exclude ordinary references.

A2.0 prompts must mention every required item: @图1 or @图片1, @视频1 and @音频1, starting at 1 without leading zeroes. Missing mentions may return 422. C audio uses @Audio1, @Audio2. Check numbering after changing material order.

The examples below show an image, frame pair, video plus audio and C audio-only input. Replace YOUR_MODEL with the matching model. All media URLs must be readable by the service.

{
  "model": "YOUR_MODEL",
  "prompt": "Animate @图1 with a smooth camera movement",
  "seconds": 8,
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "references": [
    {
      "type": "image",
      "role": "reference",
      "source": "YOUR_IMAGE_URL"
    }
  ]
}

{
  "model": "YOUR_MODEL",
  "prompt": "Move smoothly from @图1 to @图2",
  "seconds": 8,
  "resolution": "720p",
  "aspect_ratio": "auto",
  "references": [
    {
      "type": "image",
      "role": "first_frame",
      "source": "YOUR_FIRST_FRAME_URL"
    },
    {
      "type": "image",
      "role": "last_frame",
      "source": "YOUR_LAST_FRAME_URL"
    }
  ]
}

{
  "model": "YOUR_MODEL",
  "prompt": "Follow the movement in @视频1 and rhythm in @音频1",
  "seconds": 8,
  "resolution": "720p",
  "references": [
    {
      "type": "video",
      "role": "reference",
      "source": "YOUR_VIDEO_URL"
    },
    {
      "type": "audio",
      "role": "reference",
      "source": "YOUR_AUDIO_URL"
    }
  ]
}

{
  "model": "YOUR_MODEL",
  "prompt": "Create a scene following @Audio1",
  "seconds": 10,
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "generate_audio": true,
  "references": [
    {
      "type": "audio",
      "role": "reference_audio",
      "source": "YOUR_AUDIO_URL"
    }
  ]
}

Tasks, errors and costs

Poll every 3–5 seconds in general and 10–15 seconds for C. Keep the PopGo id; wait for queued/in_progress, download after completed, inspect failed. POST is not idempotent: inspect task records after a create timeout before resubmitting. See Video task workflow for downloads and backoff.

Error or symptom	Correction
400 source URL invalid / unsupported protocol: data	Use a permitted public source for A/C; A requires HTTPS
400 count, duration, resolution or mode mismatch	Check this model's contract instead of copying a sibling
422 reference absent from prompt	Fix @ numbering and mention each required reference
Download failure, 403/expiry, sharing page	Replace with a directly downloadable URL; browser preview does not prove service access
Result ignores reference	Inspect Canvas validation, PopGo parsing, outgoing fields, service acceptance and final frames separately

Billing may be per request, per second or by specification multiplier, according to current PopGo pricing and NewAPI task billing. Neither the model name nor a missing usage field establishes the price. HTTP 200 and a task ID prove only their observed stages. This feature has source, mock and offline-browser checks; real generation quality remains separately unverified.

Models covered​

Capability matrix​

Modes and combinations​

Sources, files and reference duration​

Fields, defaults and differences​

Unified references and examples​

Tasks, errors and costs​