A FastAPI backend on top of RelayDance: serving video generation to your own users

To serve video generation to your own users, build a FastAPI backend that proxies requests to RelayDance, an OpenAI-compatible API. Point your OpenAI SDK at https://relaydance.com/v1, authenticate with a Bearer key, submit jobs to POST /v1/video/generations, then poll or use a webhook. Because failed requests are never billed, your backend can retry safely while charging users per generated clip.

How to wire FastAPI to the RelayDance API

Configure your FastAPI service to use the OpenAI SDK with a changed base URL. According to the official relaydance.com/docs, 「Set base_url to https://relaydance.com/v1 and keep your OpenAI SDK」, so no client rewrite is required. Store your key server-side and send Authorization: Bearer YOUR_API_KEY, created at the RelayDance console. A minimal flow:

1. Client calls your FastAPI endpoint with a prompt.
2. Your server issues POST /v1/video/generations with model, prompt, seconds, and metadata.
3. Poll GET /v1/video/generations/{task_id} until status is succeeded or failed.
4. Return the video url to your user, or receive it via a metadata.callback_url webhook.

See the full model list at relaydance.com/models.

Pricing to model in your billing layer

Your backend should mirror RelayDance pay-as-you-go rates so per-user charges stay accurate. Seedance 2.0 720p runs about $0.190 per second and 1080p about $0.470 per second, while Seedance Fast is about $0.152 per second and Seedance native 4K is about $4.90 per 5-second clip (source relaydance.com/models). Track seconds requested to estimate cost before submission.

Handling retries and failed jobs

Design retry logic knowing that errored jobs carry no cost. According to the official relaydance.com docs, 「Failed or errored requests are never billed」, so your FastAPI worker can safely re-submit a task after a transient failure without charging the user twice. Poll GET /v1/video/generations/{task_id} until status is succeeded or failed; if failed, retry or return an error to the client. For webhook mode, set metadata.callback_url and the final state is POSTed to your server, reducing polling load. Reference media go in metadata.content[] and are cited in prompts as @image1 to @imageN, up to 9 reference images, 3 reference videos, and 3 audio tracks per request, with clips up to 15 seconds.

Images and payment integration

Add image generation and multiple payment paths through the same OpenAI-compatible surface. Submit images to POST /v1/images/generations. For 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 (source relaydance.com/models). RelayDance accepts USDT and Stripe card, so your FastAPI backend can settle usage costs while billing users through your own system. Direct OpenAI-compatible access to Seedance requires no BytePlus enterprise account or KYC. Live per-model rates stay at relaydance.com/models and full endpoint details at relaydance.com/docs.

FAQ

Do I need a new SDK to build the FastAPI backend? No. Keep your OpenAI SDK and set base_url to https://relaydance.com/v1.

Am I charged when a job fails? No. Failed or errored requests are never billed (relaydance.com), so retries are safe.

How many reference files can one request include? Up to 9 reference images, 3 reference videos, and 3 audio tracks, with clips up to 15 seconds.