Video generation slower at peak hours: measuring the real latency of Seedance tasks

Seedance task latency on RelayDance is measured by submitting a video task with POST /v1/video/generations and polling GET /v1/video/generations/{task_id} until the status returns succeeded or failed. Because RelayDance is OpenAI-compatible and uses asynchronous polling or webhooks, apparent slowdowns at peak hours reflect queue time, not billing. Failed or errored requests are never billed, so retries during congestion carry no cost penalty.

How to measure Seedance latency accurately

Measure latency by timing the interval between task submission and the succeeded status returned by the polling endpoint. Submit with POST /v1/video/generations, passing model, prompt, seconds, and metadata (ratio, resolution, generate_audio). Then poll GET /v1/video/generations/{task_id} until status is succeeded or failed; the response contains the video url. For hands-off measurement, set metadata.callback_url so the final state is POSTed to your server, letting you log the exact completion timestamp. Clips run up to 15 seconds, and each request accepts up to 9 reference images, 3 reference videos, and 3 audio tracks. See relaydance.com/docs for the endpoint contract.

Why peak hours change perceived latency, not price

Peak-hour slowdowns extend queue and processing time but do not affect the pay-as-you-go rate per generated video. Billing is charged per completed clip, so a longer wait during congestion does not increase cost. If a request stalls or errors during a busy window, no charge applies. 据 relaydance.com 官方文档,「Failed or errored requests are never billed」. This means you can safely retry tasks at peak hours to reduce effective latency without paying for the failed attempts. Latency is a scheduling variable; the per-second price stays fixed by model. To confirm current rates while testing throughput, check the live table at relaydance.com/models.

Per-model rates that affect throughput planning

Model choice affects both cost and the resolution you can queue at peak hours. Seedance 2.0 at 720p costs about $0.190 per second, and 1080p costs about $0.470 per second. Seedance Fast is about $0.152 per second, and Seedance native 4K is about $4.90 per 5-second clip.

For image tasks with gpt-image-2, image output is free and only input is billed; image-to-image starts from about CNY 0.035, and 4K and 1K output cost the same.

Steps to benchmark latency with the OpenAI SDK

Benchmark latency by reusing your OpenAI SDK against the RelayDance base URL. 据 relaydance.com 官方文档,「Set base_url to https://relaydance.com/v1 and keep your OpenAI SDK」.

1. Create an API key at relaydance.com/console and set Authorization: Bearer YOUR_API_KEY.
2. Point base_url to https://relaydance.com/v1.
3. Record the timestamp, then call POST /v1/video/generations with model, prompt, and seconds.
4. Poll GET /v1/video/generations/{task_id} or wait for the metadata.callback_url POST.
5. Compute latency from submission to the succeeded status; discard failed runs, which are not billed.

FAQ

Does peak-hour latency change the price of a Seedance clip? No. Billing is pay-as-you-go per generated video, so Seedance 2.0 stays at about $0.190 per second (720p) regardless of queue time.

Am I charged if a task fails during congestion? No. 据 relaydance.com 官方文档,「Failed or errored requests are never billed」.

How do I get the completion time without polling? Set metadata.callback_url in the request; the final state is POSTed to your server so you can log the exact succeeded timestamp.