Seedance 2.5 API
AI VideoSeedance 2.5 is ByteDance's flagship video route on OmniaKey: the provider ID dreamina-seedance-2-5-260628 serves text to video, a first frame and reference images, and every clip comes back with an AI-generated soundtrack.
- Per second of output$0.1133–$0.6259
- OperationsText to video
- Image to video
- Reference to video
- View tool pricing and billing rules
Your result will appear here once you run the model.
01
Where it fits
Seedance 2.5 is ByteDance's flagship video route on OmniaKey: the provider ID dreamina-seedance-2-5-260628 serves text to video, a first frame and reference images, and every clip comes back with an AI-generated soundtrack.
Choose Seedance 2.5 when a shot needs the strongest motion and material fidelity in the family, at $0.11 to $0.63 per second across 480P to 1080P; it is also the slowest tier here, since a 4-second 480P clip took about four minutes in our record.
02
What it does
Text to video
dreamina-seedance-2-5-260628
Generate a short video clip from a text prompt.
Image to video
dreamina-seedance-2-5-260628
Animate a still image into a short video clip.
Reference to video
dreamina-seedance-2-5-260628
Guide a clip with reference media to keep a subject, style or motion.
03
In depth
Seedance 2.5 is ByteDance's flagship video model in the Seedance family on OmniaKey, listed as dreamina-seedance-2-5-260628. One id serves text-to-video, a first frame and reference-driven shots. Above the three 2.0 tiers, it carries the widest envelope in the family: resolutions from 480P to 1080P and a single job that can run up to 30 seconds.
Reach for this id when one shot has to hold a character, a product or a set across more than fifteen seconds, or when the deliverable is 1080P. Seedance 2.0 covers the same three operations but stops at 15 seconds; the Fast and Mini tiers go no higher than 720P. Kling 3.0 is the other candidate for fixed lengths, and it has no reference operation at all.
Our October 2026 session ran six 720P prompts through this id plus one still-driven clip, and every job came back as a 24 fps MP4 with sound on the first attempt. The observations below record how long each job took and how an adaptive-ratio image job kept the still's own framing.
Long single-take product films
Give one job a whole 20-second beat sheet instead of stitching three shots: this is the only Seedance id that accepts a duration past 15 seconds and the only one priced at 1080P. Time each beat with timestamp lines, then repeat the constraints in a closing sentence.
Reference-locked character shots
Send one or more images through input.media as reference_image and the model builds the shot around them. Our reference probe failed when the payload carried audio only, so always attach at least one image or video. Reuse the same reference file across takes to keep a face stable.
Animating a still frame
Hand over a still as first_frame, set ratio to adaptive, and describe only motion, camera and sound. Our 2K still returned as a 1286×716 clip in the still's own shape, so compose the source in the aspect ratio you want to ship before spending a render.
04
Hands-on notes
All six Seedance 2.5 clips we prompted at 720P on 26 October 2026 succeeded on the first try as H.264 MP4 files at 24 fps and 5.04 seconds.
Read the testGeneration of those six Seedance 2.5 clips took between 153.6 and 252.0 seconds each, measured from request to finished file.
Read the testOur one Seedance 2.5 image-to-video clip, sent with an adaptive ratio from a Nano Banana 2.1 still, came back at 1286×716, the still's own shape.
Read the testOur 14 September 2026 smoke run rendered a 4-second 480p Seedance 2.5 text-to-video job as an 854x480 H.264 MP4 of 4.04 seconds.
A Seedance 2.5 reference job we sent with only an audio reference and no image or video failed rather than rendering.
05
Limits to plan around
- Every clip renders an audio track unless parameters.generate_audio is false, and the service turns that flag on by default. A silent deliverable needs either the setting or a strip pass in ffmpeg after the download.
- This is the most expensive Seedance id per output second at each resolution the family shares, and its 1080P cell is the highest rate across the four Seedance ids. Draft at 480P, then re-render the winners.
- Timed beats drift: a 0s-2s segment in our latte clip ran to about three seconds. Leave a second of slack per beat and name the camera angle inside each segment rather than once at the top.
- Copy the full id from the model page. The shorter aliases are not served on this deployment and answer with an error, and model ids are case-sensitive.
06
Compare nearby models
| Model | Best for | Operations | Listed price |
|---|---|---|---|
| Seedance 2.5This page dreamina-seedance-2-5-260628 | Long 1080P takes and reference-driven shots that the 2.0 tiers cannot reach. | Text to video · Image to video · Reference to video | $0.1133–$0.6259 per second |
| Seedance 2.0 dreamina-seedance-2-0-260128 | Fifteen-second 1080P work at a lower per-second rate than 2.5. | Text to video · Image to video · Reference to video | $0.077–$0.407 per second |
| Kling 3.0 kling-3.0 | Fixed 5, 10 or 15 second jobs where the charge is known before rendering. | Text to video · Image to video | $0.80–$3.00 per call |
07
Sources and verification
Guide updated 2026-10-27.
08
Call it
Media models run asynchronously. Create a task, poll it until it finishes, then download the signed asset URL.
# 1. Create the job. 202 Accepted; the body carries the task id,
# the pre-charged price, and a poll_url. Reuse one Idempotency-Key
# across retries so a timeout never starts a second paid job.
curl --fail-with-body https://api.omniakey.com/v1/media/tasks \
-H "Authorization: Bearer $OMNIAKEY_API_KEY" \
-H "Content-Type: application/json" \
-H "Idempotency-Key: dreamina-seedance-2-5-260628-001" \
-d '{
"model": "dreamina-seedance-2-5-260628",
"type": "video",
"operation": "text_to_video",
"input": {"prompt": "a paper boat on a quiet lake at sunset"},
"parameters": {"duration": 6, "resolution": "720p"}
}'
# 2. The same model id and the same poll loop serve the other operations:
# image_to_video media[] type "first_frame"
# reference_to_video media[] type "reference_image"
# Reference media travel alongside the prompt:
# {"prompt": "...", "media": [{"type": "first_frame", "url": "https://example.com/first-frame.png"}]}
# A public HTTPS URL or a base64 data URI both work.
# 3. Poll GET /v1/media/tasks/{id} until status is
# succeeded | failed | canceled. Progress carries percent and stage.
curl --fail-with-body https://api.omniakey.com/v1/media/tasks/TASK_ID \
-H "Authorization: Bearer $OMNIAKEY_API_KEY"
# 4. Download assets[0].url before expires_at, or poll again for a
# fresh signed URL. Failed and canceled jobs are refunded.
curl -sS -o out.mp4 "$(jq -r '.assets[0].url' result.json)"09
Pricing
Video jobs are billed per second of rendered output, not per token. Failed or canceled jobs are refunded automatically.
$0.1133–$0.6259per second
- 480P
- $0.1133per second
- 720P
- $0.2541per second
- 1080P
- $0.6259per second
More models from ByteDance
Related articles
Seedance 2.5 Prompt Guide: Structure, Examples and Real Clips
Six prompts, six real clips: what Seedance 2.5 followed, what it changed and how to fix it.
Read articleKling 3.0 vs Seedance 2.5
Both models are live on OmniaKey and both can serve the same brief. One is billed once per generation from a mode and duration matrix, the other per output second by resolution — and only one of them accepts a reference-driven shot.
Read articleBest AI video API models in 2026
There is no single best video model. The right route depends on whether you need a long shot, a faithful first frame, reference images, predictable cost, or a faster production queue.
Read articleSeedance 2.5 review
The meaningful upgrade is 30-second storytelling and production control, not the unverified 4K headlines circulating around it.
Read articleFrequently asked questions
How do I call Seedance 2.5 on OmniaKey?
POST to /v1/media/tasks with the model id, type "video", and operation "text_to_video" — or "image_to_video" with a first frame in input.media[] with type "first_frame". Reuse one Idempotency-Key across retries so a timeout never starts a second paid job.
What does Seedance 2.5 cost?
Seedance 2.5 bills $0.1133–$0.6259 per second of rendered video. Creation pre-charges the requested duration up front from your prepaid balance, and failed or canceled jobs are refunded automatically.
Why does the API return a task id instead of the video?
Video renders take from about a minute to several minutes, so OmniaKey runs every video route through one asynchronous lifecycle. Poll GET /v1/media/tasks/{id} until the status is succeeded, then read the signed asset URL.
How do I call Seedance 2.5 through the OmniaKey API?
Send POST https://api.omniakey.com/v1/media/tasks with an Idempotency-Key header, model dreamina-seedance-2-5-260628, type video, and an operation of text_to_video, image_to_video or reference_to_video. Poll GET /v1/media/tasks/{id} until it succeeds, then download assets[].url before the signed link expires.
How long can one Seedance 2.5 clip run?
Up to 30 seconds per job, with duration accepted from 4 to 30 seconds, or -1 to let the model choose. The 2.0, Fast and Mini ids stop at 15 seconds, so a 25-second shot rules them out. Our own clips ran at 5 seconds and returned 5.04-second files.
Does Seedance 2.5 generate its own soundtrack?
Yes. generate_audio defaults to true, and the model writes voice, effects and music to match the prompt and picture. Our first prompt said nothing about sound and came back with an instrumental bed, so add “No BGM” and “No subtitles” when you need clean effects only.
What separates Seedance 2.5 from Seedance 2.0?
Duration, resolution range and prompt syntax. 2.5 accepts 4 to 30 seconds and prices three resolutions up to 1080P, while 2.0 stops at 15 seconds. ByteDance's guide also notes that 2.0 ignores timestamps and follows shot numbers, so a timed prompt written for 2.5 needs a rewrite.
Can Seedance 2.5 build a shot from reference images or a reference video?
Yes, through the reference_to_video operation, with references supplied in input.media as reference_image entries. Audio alone is not enough: our probe with an audio reference and no image or video failed rather than rendering. Pair narration or ambience with at least one image or video reference.
