Seedance API quickstart: your first AI video in five minutes

An account, an API key and a token balance. Everything else is a plain HTTPS JSON API at https://kreatorflow.ai/v1 with the key sent as "Authorization: Bearer <key>". There is no SDK to install; curl, fetch or requests are enough. KreatorFlow is an official BytePlus partner, and the video models are ByteDance's Seedance family.

Keys carry scopes. For video you need videos:create to submit and jobs:read to poll and download; images:create, audio:create, uploads:create and models:read cover the other endpoints. A request without the right scope returns 403 scope_missing.

One POST to /v1/videos. The only required fields are model and prompt (up to 16,000 characters for Seedance 2.0); duration, aspect_ratio (ratio also works), resolution and generate_audio shape the clip. Seedance 2.0 takes 4, 5, 6, 8, 10, 12 and 15 seconds. Seedance 2.5 goes up to 30 seconds in one take.

The response is JSON with success: true and data.job (its id and status), plus balance_tokens and quoted_token_cost. The 8-second 720p clip below is quoted at 18 tokens ($0.90).

The script below creates a 5-second vertical clip, polls every 10 seconds, waits out a 429 using the Retry-After header, and writes clip.mp4 when the job completes. It runs on Node 18 or later, which has fetch built in.

The same flow with the requests library. In production, swap the polling loop for a webhook so you are not holding a worker open while the clip renders.

created and queued mean the job is accepted and waiting; processing means the model is rendering; completed means the file is ready at /v1/jobs/{id}/output. failed, canceled and blocked are final: the job carries errorCode and errorMessage, and tokens reserved for a job that did not complete are released.

Treat every final status as an answer. Resending a failed job with the same Idempotency-Key returns the same failed job, so send it again with a new key and, if the error points at the prompt or a reference, change the input first.

Errors come back as JSON with success: false, a machine-readable code and a readable error. The ones worth handling explicitly:

Networks drop responses. If your create call times out and you send it again with the same Idempotency-Key, KreatorFlow returns the job it already created (marked idempotent_replay: true) instead of starting and charging a second one. Use a fresh UUID per intended clip and reuse it only when resending that same request.

Public HTTPS URLs can go straight into the request (reference_image_urls, reference_video_urls, reference_audio_urls, or a content array). For your own files, upload first: POST /v1/uploads returns a direct_upload_url, you PUT the bytes there, then POST /v1/uploads/complete and use the returned upload id. Limits are 50 MB for images, 200 MB for video and 15 MB for audio. Seedance 2.0 takes up to 9 images, 3 clips and 3 audio files per job.

A reference video's seconds are billed too: input plus output seconds at 60% of the per-second rate, and never less than the same job without the video. first_frame_upload_id and last_frame_upload_id pin the opening and closing frame.

Video is billed per second of output at a rate set by model and resolution, rounded up to the cent and then to the next whole token. As of 17 October 2026: Seedance 2.0 is $0.110 per second at 720p and $0.240 at 1080p. Seedance 2.0 Fast is $0.084 per second at 720p, for drafts. The full table is below; the create response always carries the exact quote.