Seedance 2.5 4K API

Native 4K video from text or images, with audio.

~192.31s

POST /v2/seedance-2.5-4k · submit + poll

 1# pip install "segmind>=1.1.0"
 2# export SEGMIND_API_KEY="YOUR_API_KEY"
 3from segmind import SegmindClient, InferenceFailed, InferenceTimeout
 4
 5# Async (v2) — recommended for long-running / video models.
 6# run() blocks up to 600s; submit_async + job.wait(timeout=...) sets a longer
 7# deadline and keeps the request_id so you can re-poll later.
 8client = SegmindClient()                      # reads SEGMIND_API_KEY
 9payload = {
10    "prompt": "A ruby-throated hummingbird hovers at an orange trumpet flower in a shaded garden, iridescent green feathers and fast-beating wings crisp in extreme close-up; it slips its slender beak into the bloom, knocks a puff of pollen into the air, then darts upward out of frame. Steady macro shot. Natural sound of humming wingbeats and rustling leaves.",
11    "first_frame_url": "",
12    "last_frame_url": "",
13    "reference_images": [],
14    "reference_audios": [],
15    "duration": 4,
16    "resolution": "4k",
17    "aspect_ratio": "16:9",
18    "generate_audio": True,
19    "seed": 42,
20    "return_last_frame": False,
21    "skip_moderation": False,
22    "bitrate_mode": "standard",
23    "auto_resize": True,
24    "output_codec": "h264",
25}
26job = client.submit_async("seedance-2.5-4k", **payload)
27print(job.request_id)                         # available immediately
28try:
29    result = job.wait(timeout=900, interval=2.0)
30    print(result["status"])                  # COMPLETED
31    print(result.get("output"))              # model output (e.g. video URL)
32except InferenceTimeout as e:
33    print("still running:", e.request_id)    # re-poll later with this id
34except InferenceFailed as e:
35    print("failed:", e.detail)
36
37# Fast models (<=600s) can use the one-liner instead:
38# result = segmind.run("seedance-2.5-4k", **payload)

 1# pip install "segmind>=1.1.0"
 2# export SEGMIND_API_KEY="YOUR_API_KEY"
 3from segmind import SegmindClient, InferenceFailed, InferenceTimeout
 4
 5# Async (v2) — recommended for long-running / video models.
 6# run() blocks up to 600s; submit_async + job.wait(timeout=...) sets a longer
 7# deadline and keeps the request_id so you can re-poll later.
 8client = SegmindClient()                      # reads SEGMIND_API_KEY
 9payload = {
10    "prompt": "A ruby-throated hummingbird hovers at an orange trumpet flower in a shaded garden, iridescent green feathers and fast-beating wings crisp in extreme close-up; it slips its slender beak into the bloom, knocks a puff of pollen into the air, then darts upward out of frame. Steady macro shot. Natural sound of humming wingbeats and rustling leaves.",
11    "first_frame_url": "",
12    "last_frame_url": "",
13    "reference_images": [],
14    "reference_audios": [],
15    "duration": 4,
16    "resolution": "4k",
17    "aspect_ratio": "16:9",
18    "generate_audio": True,
19    "seed": 42,
20    "return_last_frame": False,
21    "skip_moderation": False,
22    "bitrate_mode": "standard",
23    "auto_resize": True,
24    "output_codec": "h264",
25}
26job = client.submit_async("seedance-2.5-4k", **payload)
27print(job.request_id)                         # available immediately
28try:
29    result = job.wait(timeout=900, interval=2.0)
30    print(result["status"])                  # COMPLETED
31    print(result.get("output"))              # model output (e.g. video URL)
32except InferenceTimeout as e:
33    print("still running:", e.request_id)    # re-poll later with this id
34except InferenceFailed as e:
35    print("failed:", e.detail)
36
37# Fast models (<=600s) can use the one-liner instead:
38# result = segmind.run("seedance-2.5-4k", **payload)

API Endpoint

POSThttps://api.segmind.com/v1/seedance-2.5-4k

Parameters

promptrequired

string

Text describing the video. Use Shot 1:, Shot 2: for multi-shot sequences. Reference uploaded assets as '@Image 1', '@Video 1', '@Audio 1' in your prompt.

aspect_ratiooptional

string

Output aspect ratio. Defaults to adaptive, which lets the model match the prompt or the input media. adaptive is the only value accepted when you supply first_frame_url, or when the prompt asks for a video edit or extension — the output keeps the geometry of the input in those cases. Pick a fixed ratio (16:9 landscape, 9:16 vertical, 21:9 ultrawide) for text-to-video or plain reference-to-video.

Default: "adaptive"

Allowed values :

"16:9""9:16""1:1""4:3""3:4""21:9""adaptive"

auto_resizeoptional

boolean

Automatically resizes an input image that falls outside the provider's accepted range (300-6000px per side, aspect ratio 0.4-2.5, 30MB) so the request succeeds instead of being rejected. Aspect ratio is preserved: oversized images are downscaled, undersized ones upscaled, and out-of-band ratios padded rather than cropped. On by default - set it to false to have the provider reject out-of-range images instead. When a resize happens the response carries an X-Input-Normalized header naming the field and the before/after dimensions.

Default: true

bitrate_modeoptional

string

Output video encoding bitrate. 'standard' gives smaller files and faster downloads; 'high' produces ~5-6x higher bitrate for better visual fidelity and fewer compression artifacts (larger files). Does not affect price.

Default: "standard"

Allowed values :

Standard→"standard"

High→"high"

durationoptional

integer

Video length in seconds (4-30), or -1 to let the model choose. Video-editing prompts require -1 (output follows the input video, which must be 4-30s). Cost scales with output duration (token-based).

Default: 5

Allowed values (28 total):

Auto (use for video edits)→-1

4s→4

5s→5

6s→6

7s→7

8s→8

9s→9

10s→10

11s→11

12s→12

+18 more

first_frame_urloptional

string (uri)

Starting frame image URL for image-to-video. Animates outward from this reference. Note: images with real human faces are blocked by ByteDance content policy. Cannot be used together with reference_images.

generate_audiooptional

boolean

Co-generate synchronized audio (dialogue, SFX, ambient, music) in the same pass. Best for scenes with speech, nature sounds, or music.

Default: true

last_frame_urloptional

string (uri)

Ending frame; requires first_frame_url. Guides transitions between two frames.

output_codecoptional

string

auto (default) returns the video exactly as the model encodes it: 10-bit HEVC at 4K. h264 re-encodes it for browsers that cannot play HEVC; at 4K that adds noticeable time.

Default: "auto"

Allowed values :

"auto""h264"

reference_audiosoptional

string[]

Up to 10 reference audio files (MP3, 2-30 seconds each, 30 seconds combined, max 15MB each). Cite as '@Audio 1' in your prompt. Audio-only reference is supported on seedance-2.5.

reference_imagesoptional

string[]

Up to 30 reference images (max 30MB each) for character/style consistency. Cite as '@Image 1', '@Image 2' in your prompt. Cannot be used together with first_frame_url.

reference_videosoptional

string[]

Up to 10 reference videos (2-30 seconds each, 30 seconds combined, mp4/mov, max 200MB each) for motion/style transfer. Cite as '@Video 1' in your prompt. Each clip's full length is billed at the video-input token rate, counted at the output's 4K resolution, on top of the output.

resolutionoptional

string

Output resolution. This model renders natively at 4K only (3840x2160 at 16:9, 10-bit, 24fps). Use seedance-2.5 for 480p, 720p or 1080p.

Default: "4k"

Allowed values :

4K→"4k"

return_last_frameoptional

boolean

Returns the final video frame as a separate image, delivered in the response as `video.last_frame_url`. Chain shots by feeding that URL into the next generation's `first_frame_url`. Note this is an output: it is distinct from the `last_frame_url` input parameter, which is an image you supply to steer how the clip ends.

Default: false

seedoptional

integer

Reproducibility seed. Use -1 for random output, set a fixed value to iterate on the same scene.

Default: 42Range: -1 - 2147483647

skip_moderationoptional

boolean

Bypass the upstream content moderation pre-filter. Useful when generating artistic content or faces that may trigger false positives. Baseline safety policies still apply.

Default: true

Response Type

Returns: Video

Asynchronous requests (v2)

Use Async for video, long-running (>~60s), or high-concurrency workloads; Sync is simplest for fast image & LLM calls. Async submits a request and you poll it to completion.

1
POST /v2/seedance-2.5-4k
Submit — returns request_id, status_url, response_url
2
GET /v2/requests/{id}/status
Poll — until COMPLETED or FAILED
3
GET /v2/requests/{id}
Result — final response body

Status states

QUEUED— Accepted, waiting for a worker

PROCESSING— Running on a worker

COMPLETED— Done — result body is ready

FAILED— Errored (incl. content/RAI blocks)

A FAILED request is served as HTTP 422 — the body still carries the error detail.
An unknown or expired request_id returns HTTP 404.
Results are retained for 1 hour, then expire.
Content / RAI blocks surface as FAILED, not a separate state.
Track completion by polling the status endpoint.

Common Error Codes

The API returns standard HTTP status codes. Detailed error messages are provided in the response body.

400

Bad Request

Invalid parameters or request format

401

Unauthorized

Missing or invalid API key

403

Forbidden

Insufficient permissions

404

Not Found

Model or endpoint not found

406

Insufficient Credits

Not enough credits to process request

429

Rate Limited

Too many requests

500

Server Error

Internal server error

502

Bad Gateway

Service temporarily unavailable

504

Timeout

Request timed out