ByteDance Seedseedance-2.5-flash
Seedance 2.5 at about a third less per second: 4–30 second clips at 480p or 720p, up to 20 reference assets and native support for real-person references, no manual approval or extra verification needed
$0.0572/ second-15%$0.0673official priceNo charge for failed generations
- 4–30 second clips
- 480p and 720p
- Up to 10 images, 5 videos, 5 audios
- Audio-only reference accepted
- Edits videos of 4–30 seconds
- Modality
- Video
- Endpoint
- /v1/videos/generations
- Model ID
- seedance-2.5-flash
- Provider
- ByteDance Seed
Seedance 2.5 Flash API overview
Documented endpoint
Video requests go to /v1/videos/generations. The parameter table below lists the fields this surface documents.
Model provider
Seedance 2.5 Flash is published by ByteDance Seed and is called through the API model ID seedance-2.5-flash.
Output and limits
4–30 seconds; 480p or 720p; one output per request; reference videos add their seconds to the billed seconds; edits videos of 4–30 seconds; no drafts
Seedance 2.5 Flash API pricing
Seedance 2.5 Flash is billed 15% below the official list price on every metered dimension below.
| Item | ArgoLink rate | Official price | You save |
|---|---|---|---|
| 480p | $0.0572 / second | -15% | |
| 720p | $0.129 / second | -15% | |
| 480p with video input | $0.0351 / second | -15% | |
| 720p with video input | $0.0792 / second | -15% |
4–30 seconds; 480p or 720p; one output per request; reference videos add their seconds to the billed seconds; edits videos of 4–30 seconds; no drafts
Compared with other platforms
| Platform | 480p / second | 720p / second |
|---|---|---|
| ArgoLink | $0.0572 | $0.129 |
| Official | $0.0673 -15% | $0.1512 -15% |
| fal | $0.1442 -60% | $0.3094 -58% |
Each platform's list price for a request without video input, as its own page showed it. The green figure is how much less ArgoLink charges. Checked on
Seedance 2.5 Flash API reference
Quick start
Create an API key in the console, then use the Base URL, endpoint, and request body shown on this page.
curl https://argolink.io/v1/videos/generations \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.5-flash",
"prompt": "A slow aerial shot over a foggy pine forest at sunrise"
}'
curl https://argolink.io/v1/videos/$REQUEST_ID \
-H "Authorization: Bearer $ARGOLINK_API_KEY"
curl -L https://argolink.io/v1/videos/$REQUEST_ID/content \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-o output.mp4
import os
import time
import requests
headers = {"Authorization": f"Bearer {os.environ['ARGOLINK_API_KEY']}"}
job = requests.post(
"https://argolink.io/v1/videos/generations",
headers=headers,
json={"model": "seedance-2.5-flash", "prompt": "A slow aerial shot over a foggy pine forest at sunrise"},
timeout=600,
)
job.raise_for_status()
request_id = job.json()["request_id"]
while True:
status = requests.get(f"https://argolink.io/v1/videos/{request_id}", headers=headers, timeout=60).json()
if status["status"] in ("done", "failed", "expired"):
break
time.sleep(5)
if status["status"] == "done":
video = requests.get(f"https://argolink.io/v1/videos/{request_id}/content", headers=headers, timeout=600)
open("output.mp4", "wb").write(video.content)
Modes
Text to video
Nothing but a prompt
The shortest call there is: only prompt is required, and the fields you leave out render and bill 5 seconds at 720p 16:9.
curl https://argolink.io/v1/videos/generations \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.5-flash",
"prompt": "A slow aerial shot over a foggy pine forest at sunrise",
"duration": 5,
"resolution": "720p",
"aspect_ratio": "16:9"
}'
| Field | Value |
|---|---|
prompt | required, up to 40,000 bytes; describe the camera move, the subject and the mood |
duration | 4–30 whole seconds, defaults to 5 |
resolution | 480p · 720p, defaults to 720p |
aspect_ratio | 21:9 · 16:9 · 4:3 · 1:1 · 3:4 · 9:16, defaults to 16:9 |
size | pixels instead of resolution plus aspect_ratio, so 1280x720 is 720p 16:9 |
One request renders one video; seed and watermark are not supported, and the result carries a generated soundtrack unless generate_audio is false.
Reference
Follow images, video or audio you supply
The result follows the subject, motion or beat of the assets you send. Point at one from the prompt with @Image 1, @Video 1 or @Audio 1; assets you never name are still used.
curl https://argolink.io/v1/videos/generations \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.5-flash",
"prompt": "Let @Image 1 walk into a rainy street at night",
"duration": 5,
"resolution": "720p",
"reference_images": [{"url": "https://example.com/subject.png"}]
}'
curl https://argolink.io/v1/videos/generations \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.5-flash",
"prompt": "The girl in @Image 1 dances to the beat of @Audio 1, moving like the dancer in @Video 1",
"duration": 10,
"resolution": "720p",
"reference_images": [{"url": "https://example.com/girl.png"}],
"reference_videos": [{"url": "https://example.com/dance.mp4"}],
"reference_audios": [{"url": "https://example.com/beat.mp3"}]
}'
| Field | Value |
|---|---|
prompt | required; name an asset as @Image 1, @Video 1 or @Audio 1 |
reference_images | up to 10 HTTPS images, JPEG or PNG |
reference_videos | up to 5 clips, each 2–30 seconds and 30 seconds together |
reference_audios | up to 5 clips, each 2–30 seconds and 30 seconds together |
duration | 4–30 whole seconds, defaults to 5 |
resolution | 480p · 720p, defaults to 720p |
aspect_ratio | 21:9 · 16:9 · 4:3 · 1:1 · 3:4 · 9:16, defaults to 16:9 |
The three kinds mix freely, up to 20 assets in total. Reference videos are billed at the output rate; reference images and audio are free.
First and last frames
Pin the opening, and optionally the close
start_image alone is a first frame; add end_image and the model fills the middle between the two.
curl https://argolink.io/v1/videos/generations \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.5-flash",
"prompt": "The camera pushes in slowly while leaves drift across the frame",
"duration": 5,
"resolution": "720p",
"aspect_ratio": "adaptive",
"start_image": {"url": "https://example.com/first-frame.png"}
}'
curl https://argolink.io/v1/videos/generations \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.5-flash",
"prompt": "Day turns into night over the same skyline as the lights come on",
"duration": 5,
"resolution": "720p",
"aspect_ratio": "adaptive",
"start_image": {"url": "https://example.com/day.png"},
"end_image": {"url": "https://example.com/night.png"}
}'
| Field | Value |
|---|---|
prompt | required; describe how the shot moves |
start_image | one HTTPS image, and it sets the frame of the result |
end_image | optional; one HTTPS image the video ends on |
duration | 4–30 whole seconds, defaults to 5 |
resolution | 480p · 720p, defaults to 720p |
aspect_ratio | adaptive only; it follows the first frame |
Neither field can be combined with any reference asset. The result takes the ratio closest to the first frame.
Edit a video
Rewrite footage you already have
The only mode that starts from a video: the result keeps that video's length and ratio, and you pick the tier with resolution.
curl https://argolink.io/v1/videos/generations \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.5-flash",
"omni_reference_task_type": "edit",
"prompt": "Make the colors warmer and keep everything else",
"resolution": "720p",
"reference_videos": [{"url": "https://example.com/street.mp4"}]
}'
| Field | Value |
|---|---|
omni_reference_task_type | edit |
reference_videos | the first clip is the video to edit, 4–30 seconds long |
reference_images | up to 10 images steer the edit |
prompt | required; say what to change |
resolution | 480p · 720p, defaults to 720p |
duration · aspect_ratio | taken from the edited video, so leave both out |
The edited video is billed as output and as a reference video: editing a 5-second video bills 10 seconds. start_image and end_image cannot be sent here.
Extend a video
Continue footage you already have
Continues the first video in reference_videos, forward or backward as the prompt says. The result is the new footage only, at the duration you ask for and in that video's aspect ratio.
curl https://argolink.io/v1/videos/generations \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "seedance-2.5-flash",
"omni_reference_task_type": "extend",
"prompt": "Extend it forward: the camera keeps rising above the forest",
"duration": 5,
"resolution": "720p",
"reference_videos": [{"url": "https://example.com/forest.mp4"}]
}'
| Field | Value |
|---|---|
omni_reference_task_type | extend |
reference_videos | the first clip is the video to continue; up to 5 clips, each 2–30 seconds and 30 seconds together |
reference_images · reference_audios | optional; up to 10 images and 5 audio clips steer the extension |
prompt | required; say which way to continue and what happens next |
duration | 4–30 whole seconds, the length of the new footage |
resolution | 480p · 720p, defaults to 720p |
aspect_ratio | taken from the video you extend, so leave it out |
Billed at the official price. The reference videos are billed along with the new footage: extending a 5-second video by 10 seconds bills 17 seconds, the official minimum.
How to upload and use reference assets
Your own link, or a local file uploaded first
The url of start_image, end_image and every entry of reference_images, reference_videos and reference_audios is an HTTPS address. A link you already host goes in as it is, as long as it downloads without signing in and every redirect stays on HTTPS. A local file is uploaded first, and the media_url you get back goes into the request.
FILE=dance.mp4
TICKET=$(curl -fsS https://argolink.io/v1/media/uploads \
-H "Authorization: Bearer $ARGOLINK_API_KEY" \
-H "Content-Type: application/json" \
-d "{\"model\":\"seedance-2.5-flash\",\"type\":\"video\",\"content_type\":\"video/mp4\",\"size_bytes\":$(wc -c < "$FILE" | tr -d ' ')}")
curl -fsS -X PUT "$(jq -r .upload_url <<<"$TICKET")" \
-H "Content-Type: video/mp4" --data-binary @"$FILE"
# Put this media_url in reference_videos[].url
jq -r .media_url <<<"$TICKET"
Asking for a ticket is free and creates no job: upload_url is valid for 15 minutes and the PUT must carry the Content-Type you declared; the media_url it exchanges for stays readable for seven days.
| Field | Format | Size | Length |
|---|---|---|---|
reference_images · start_image · end_image | JPEG or PNG | ≤ 20 MiB | — |
reference_videos | MP4 or MOV | ≤ 100 MiB | 2–30 seconds each |
reference_audios | WAV or MP3 | ≤ 15 MB | 2–30 seconds each |
reference_videos[0] (the video to edit) | MP4 or MOV | ≤ 100 MiB | 4–30 seconds |
An asset outside these ranges fails the job with invalid_input before generation starts, and nothing is charged.
How to reference your assets in the prompt
Name one asset from the prompt
Name a reference with @, its type and its number. The number is its position in reference_images, reference_videos or reference_audios, and each array is counted from 1.
| In the prompt | Refers to |
|---|---|
@Image 1 | 1st item in reference_images |
@Video 1 | 1st item in reference_videos |
@Audio 1 | 1st item in reference_audios |
{
"model": "seedance-2.5-flash",
"prompt": "The girl in @Image 1 dances to the beat of @Audio 1, moving like the dancer in @Video 1",
"reference_images": [
{
"url": "https://example.com/girl.png"
}
],
"reference_videos": [
{
"url": "https://example.com/dance.mp4"
}
],
"reference_audios": [
{
"url": "https://example.com/beat.mp3"
}
]
}
This is the spelling the console's workbench writes when you pick a reference after @. Also recognized: image 1 without @, <image 1>, @image1 and the same words in other languages, such as @图片1 or @imagen 1. Naming is optional: every reference is used either way. With @ or < >, a number past the end of its array is rejected. When editing a video (omni_reference_task_type edit), @Video 1 is the video being edited and needs no mention; the other references keep their numbers.
Seedance 2.5 Flash full request parameters
| Parameter | Type | Requirement | What it does |
|---|---|---|---|
model | string | required | The catalog identifier for this page: seedance-2.5-flash. |
prompt | string | required | Describes the video: camera motion, subject action and mood. This model needs one in every mode, assets or not. Name a reference in the prompt as @Image 1 or @Video 2. |
duration | integer | optional | Output length in whole seconds, 4 to 30; defaults to 5. An edit keeps the length of the video it edits, so leave it out there. |
resolution | string | optional | Output resolution: 480p or 720p; defaults to 720p. |
aspect_ratio | string | optional | Output aspect ratio: 21:9, 16:9, 4:3, 1:1, 3:4 or 9:16; defaults to 16:9. It applies to reference requests, with or without assets. In first-frame and first-and-last-frame modes, send adaptive only; the result follows the first frame and uses the ratio closest to its first frame. An edit keeps its video's ratio, so leave it out there too. |
size | string | optional | Pixels instead of resolution plus aspect_ratio, written WIDTHxHEIGHT: the short side picks the tier (480 for 480p or 720 for 720p) and the shape picks the ratio, so 1280x720 is 720p 16:9. It picks a tier and a ratio, not exact pixels; a resolution or aspect_ratio sent with it must match, or the request is rejected with 400. An edit takes a resolution name only. |
start_image | object | first-frame modes | HTTPS URL of the first frame. Alone it makes a first-frame video, with end_image a first-and-last-frame video; it cannot be combined with reference assets. |
end_image | object | first and last frames only | HTTPS URL of the last frame; send it with start_image. The source ratio may differ because the result adapts it to the first frame. |
reference_images | array | reference and edit | HTTPS URLs of reference images, up to 10. Images, videos and audio together are limited to 20 per request. |
reference_videos | array | reference and edit | HTTPS URLs of reference videos, up to 5 clips, each 2–30 seconds and 30 seconds together. Their seconds are billed like the output's. In an edit the first video is the one edited, 4–30 seconds long. |
reference_audios | array | reference and edit | HTTPS URLs of reference audio clips, up to 5, each 2–30 seconds and 30 seconds together. Audio can be sent on its own. |
omni_reference_task_type | string | edit or extend | Set it to edit to edit the first entry of reference_videos, 4–30 seconds long, or to extend to continue it by 4–30 seconds. Left out, auto and reference all make an ordinary request. |
seconds · ratio | — | compatible names | Aliases of duration and aspect_ratio; send one name per value, never both. |
image_url · image_urls · video_urls · audio_urls · frame_images · input_references | — | compatible names | Field names from fal and OpenRouter requests work too: image_url is start_image; image_urls, video_urls and audio_urls are the three reference lists; frame_images items whose frame_type is first_frame or last_frame are start_image and end_image; input_references items of type image_url, video_url or audio_url are references. Give each input one name. |
n | integer | optional | Must be 1 or omitted; every request produces exactly one video. |
generate_audio | boolean | optional | Defaults to true: Seedance 2.5 Flash scores the video itself. Send false for a silent result. |
seed · watermark | — | not supported | Not exposed on this model: any seed, or watermark: true, is rejected with 400. |
request_id | string | response field | Returned by the submit call and the only identifier accepted by the status and content endpoints. |
progress · estimated_seconds_remaining | integer | response field | Returned by the status endpoint while status is pending: progress estimates how far the job has come, from 0 to 99, and estimated_seconds_remaining how many seconds are left; the latter is left out until there is an estimate. progress is 100 once status is done. |
video.url | string | response field | Download link of the finished MP4 once status is done: an ArgoLink link anyone can open without the API key, for 30 days. |
usage | object | response field | Returned once status is done: output_seconds, reference_video_seconds and billed_seconds, their sum, which the job is billed for. |
Seedance 2.5 Flash API compared with similar models
| Model | Model ID | Provider | Rate | Best for |
|---|---|---|---|---|
| Seedance 2.5 Flash | seedance-2.5-flash | ByteDance Seed | $0.0572 / second | Seedance 2.5 at about a third less per second: 4–30 second clips at 480p or 720p, up to 20 reference assets and native support for real-person references, no manual approval or extra verification needed |
| Seedance 2.5 | seedance-2.5 | ByteDance Seed | $0.0874 / second | The most capable Seedance video model yet: 30-second clips, up to 50 reference assets and native support for real-person references, no manual approval or extra verification needed |
| Grok Imagine Video 1.5 | grok-imagine-video-1.5 | xAI | $0.032 / second | 480p, 720p, and 1080p video generation billed per output second |
| Gemini Omni 1.2 Flash | gemini-omni-1.2-flash | $0.0455 / second | Google Gemini Omni 1.2 Flash: 3 to 15 second clips with native sound at 720p, 1080p or 4K, from a prompt, a first and last frame, up to 10 reference images and 3 reference videos, or as an edit of a video |
More models from this provider
- Seedance 2.5The most capable Seedance video model yet: 30-second clips, up to 50 reference assets and native support for real-person references, no manual approval or extra verification needed$0.0874 / second
- Seedance 2.0The Seedance generation before 2.5: sharp clips up to 4K, driven by a prompt, frames or reference assets$0.0598 / second
- Seedance 2.0 FastThe quickest Seedance turnaround: 480p/720p clips built for previewing an idea before you commit$0.0478 / second
- Seedance 2.0 MiniThe cheapest way to run Seedance: 480p/720p clips at the lowest per-second price, made for drafts and batches$0.0299 / second
Not released yet
Other models in this modality
- Grok Imagine Video 1.5480p, 720p, and 1080p video generation billed per output second$0.032 / second
- Gemini Omni 1.2 FlashGoogle Gemini Omni 1.2 Flash: 3 to 15 second clips with native sound at 720p, 1080p or 4K, from a prompt, a first and last frame, up to 10 reference images and 3 reference videos, or as an edit of a video$0.0455 / second
- Kling 4.0Kuaishou Kling 4.0: 3 to 30 second clips at 720p or 1080p from a prompt, or from up to 10 reference images and 5 reference videos, with a soundtrack of its own$0.1413 / second
- MiniMax H3MiniMax H3: 2K video from a prompt, a first frame, a frame pair or reference assets$0.039 / second
Seedance 2.5 Flash API FAQ
How do I call Seedance 2.5 Flash?
Create an API key in the console and POST https://argolink.io/v1/videos/generations with model set to seedance-2.5-flash. The response carries request_id, the only job identifier: poll GET https://argolink.io/v1/videos/{request_id}, which reports progress and estimated_seconds_remaining while the job is pending, until status is done, then download the MP4 from GET https://argolink.io/v1/videos/{request_id}/content, the address the status returns as video.url.
How much does Seedance 2.5 Flash cost?
Per output second, from $0.0572 per second at the lowest tier; each tier in the pricing table has its own per-second rate, multiplied by the video's length. Every tier is 15% below the official per-second list price ($0.0673 per second for the lowest tier). Reference videos are billed at the same per-second rate: billed seconds are the output seconds plus the reference videos' total seconds, with any part of a second left off. A video with a reference video bills at least 5/3 of its output seconds, rounded up, which is the official minimum, so a 5-second video with a 2.4-second reference video bills 9 seconds. Reference images and audio are free. Only a completed video is billed; a failed or rejected job costs nothing. With video input, a reference video or the video an edit renders from, every billed second is charged at the with-video-input rate in the pricing table.
Which input modes does Seedance 2.5 Flash support?
All on the same endpoint: reference (a prompt alone, or with images, videos and audio as references: up to 10 reference_images, 5 reference_videos and 5 reference_audios, 20 in total, in any combination); first frame (start_image); first and last frames (start_image plus end_image on one aspect ratio); video editing (omni_reference_task_type edit, which edits the first entry of reference_videos). First-frame and first-and-last-frame videos take their aspect ratio from the first frame. Every asset is an HTTPS URL.
What does Seedance 2.5 Flash not support?
More than one video per request (n must be 1 or omitted), any seed, or watermark: true; those return 400. Output is an H.264 MP4 that carries a generated soundtrack unless generate_audio is false.
How is the aspect ratio of a Seedance 2.5 Flash first-frame video chosen?
Leave aspect_ratio out, or send adaptive, and the first frame decides: the video takes the closest of 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, so the frame is cropped as little as possible. The two frame modes take no fixed ratio; naming one is rejected. With first and last frames, the source images may have different ratios; the result adapts the last frame to the first.
Do Seedance 2.5 Flash videos have sound?
By default, yes: Seedance 2.5 Flash scores the video itself. Send generate_audio: false for a silent result.
Can Seedance 2.5 Flash use reference images with real human faces?
Yes. Seedance 2.5 Flash takes photos of real people directly as reference images: there is no liveness check, no asset library and no asset:// ID to set up first.
How is Seedance 2.5 Flash different from Seedance 2.5?
Flash renders at 480p or 720p and takes up to 10 reference images, 5 videos and 5 audio clips, at about a third less per second. Seedance 2.5 adds 1080p and 4K, larger reference sets (30 images, 10 videos, 10 audio clips) and drafts.
How long does a Seedance 2.5 Flash generation take?
Measured in production, about 2 to 4 minutes for a 4 to 10 second clip. A busy queue can occasionally hold a job much longer, so poll the job status instead of assuming a deadline.
Can Seedance 2.5 Flash return 1080p or several clips at once?
1080p, no: Flash renders at 480p or 720p, so call seedance-2.5 for 1080p or 4K. Output is a 24 fps H.264 MP4 with an AAC stereo audio track unless generate_audio is false. Several clips, no: every request produces exactly one video, so n must be 1 or omitted.