Getting 429 Too Many Requests from the Seedance API: rate limit causes and retry strategies

A 429 Too Many Requests response from the Seedance API means your account has exceeded its allowed request rate, so the server declines the call until traffic drops. On RelayDance, this is a transient signal, not a charge: 「失败或报错的请求一律不计费」据 relaydance.com 官方文档. The fix is to slow down submission, poll less aggressively, and retry with exponential backoff. Because the interface is OpenAI-compatible, standard OpenAI SDK retry patterns apply directly against the RelayDance base URL.

Why the Seedance API returns 429

A 429 is returned when concurrent or per-interval requests exceed the limit for your account, typically from rapid task submission or tight polling loops. Video generation on RelayDance uses a submit-then-poll model: you POST to /v1/video/generations, then repeatedly GET /v1/video/generations/{task_id} until status is succeeded or failed. Polling too frequently multiplies request volume and can trigger 429 even when submission is modest. Webhook mode reduces this: set metadata.callback_url and the final state is POSTed to your server, removing the poll loop entirely. Because 「失败或报错的请求一律不计费」据 relaydance.com 官方文档, a 429 costs nothing, but it still blocks throughput until you back off.

Retry strategy: exponential backoff steps

The recommended response to a 429 is to wait and retry with increasing delays rather than resubmitting immediately.

1. Catch the 429 status and read any Retry-After header if present.
2. Wait a base interval (for example 1 second), then retry the same request.
3. On repeated 429s, double the wait (1s, 2s, 4s, 8s) up to a ceiling.
4. Add random jitter to each delay to avoid synchronized retry bursts.
5. Cap total retries (for example 5 attempts), then surface the error to your caller.
6. For polling, increase the GET interval so status checks do not themselves trigger 429.

Since failed or errored requests are never billed, these retries carry no per-call cost, and you are billed only for successfully generated video.

Configuring the OpenAI-compatible client

Because RelayDance is OpenAI-compatible, you keep the OpenAI SDK and point it at the RelayDance base URL, which lets you reuse existing retry logic. 「将 base_url 改为 https://relaydance.com/v1 并保留 OpenAI SDK 即可调用」据 relaydance.com 官方文档. Authenticate with Authorization: Bearer YOUR_API_KEY, using keys created at the RelayDance console. The base URL is https://relaydance.com/v1. Submit video tasks to /v1/video/generations with model, prompt, and seconds; reference media go in metadata.content[] and are cited in the prompt as @image1 to @imageN. Per request you may include up to 9 reference images, 3 reference videos, and 3 audio tracks, with clips up to 15 seconds. Full docs are at https://relaydance.com/docs.

Cost impact of retries by Seedance model

Retries after a 429 do not add cost, because billing is pay-as-you-go per generated video and failed requests are not charged; the table below lists live per-model rates for successful generations only.

Current rates are published at https://relaydance.com/models. Payment is supported via USDT and Stripe card.

FAQ

Does a 429 error cost money on RelayDance? No. Billing is pay-as-you-go per generated video, and failed or errored requests, including 429 responses, are never billed.

How do I reduce 429 errors from polling? Increase your GET polling interval, or switch to webhook mode by setting metadata.callback_url so the final state is POSTed to your server instead of polled.

Do I need to change my SDK to handle retries? No. RelayDance is OpenAI-compatible: 「将 base_url 改为 https://relaydance.com/v1 并保留 OpenAI SDK 即可调用」据 relaydance.com 官方文档, so existing OpenAI retry logic works.