Multi reference image fusion for video generation

RelayDance supports multi reference image fusion through the metadata.content[] array on its video generation endpoint. You upload reference media, then cite each item inside your prompt using the tags @image1 to @imageN. A single request accepts up to 9 reference images, 3 reference videos and 3 audio tracks, with clips up to 15 seconds. Submit a task with POST /v1/video/generations, then poll the task or use a webhook to retrieve the final video url. RelayDance is OpenAI compatible and works with models including Seedance and Grok Imagine.

How reference fusion works

To fuse multiple references, place each media item in the metadata.content[] array of your request body. Each entry is addressable in the prompt by index: the first item is @image1, the second is @image2, and so on through @imageN. This lets the model combine subjects, styles or scene elements drawn from several inputs into one generated video. The per request limits are 9 reference images, 3 reference videos and 3 audio tracks. Reference clips can be up to 15 seconds in length. Citing references explicitly in the prompt tells the model exactly how each input should influence the output.

Submitting a fusion request

Send the task to the base URL https://relaydance.com/v1 with your API key. Authenticate using the header Authorization: Bearer YOUR_API_KEY, with keys created in the console at https://relaydance.com/console. The body uses standard fields: model, prompt, seconds, and metadata. Inside metadata you set generation options (ratio, resolution, generate_audio), an optional callback_url, and your content[] reference media. The endpoint returns a task identifier. Because RelayDance is OpenAI compatible, existing OpenAI style clients can call it by changing the base URL and key. Full request details are documented at https://relaydance.com/docs.

Steps to fuse references into a video

1. Create an API key in the console at https://relaydance.com/console.
2. Build your metadata.content[] array with up to 9 images, 3 videos and 3 audio tracks.
3. Reference each item in your prompt as @image1 through @imageN.
4. POST the request to /v1/video/generations with model, prompt, and seconds.
5. Poll GET /v1/video/generations/{task_id} until status is succeeded or failed.
6. Read the video url from the result, or receive it via webhook if you set metadata.callback_url.

Retrieving results: polling vs webhook

RelayDance offers two ways to get a completed video. With polling, you call GET /v1/video/generations/{task_id} repeatedly until the status reports succeeded or failed; the succeeded result contains the video url. With webhook mode, you set metadata.callback_url and the final state is POSTed to your server automatically. Choose based on your architecture: polling is simple for scripts, while webhooks suit production systems that should react when a task finishes.

Models, pricing and billing

Reference fusion can run on RelayDance video models including ByteDance Seedance 2.0, Seedance 1.5 Pro, HappyHorse video models, and Grok Imagine video models. RelayDance also provides Grok Imagine image models and Grok text models, plus image generation via POST /v1/images/generations. Mainstream model groups are priced 30 percent or more below official list prices, and failed or errored requests are never billed. Payment options include USDT and Stripe card. Live per model rates are published at https://relaydance.com/models, so you can confirm current costs before running fusion jobs.

FAQ

How many reference images can I use in one request? Up to 9 reference images, plus up to 3 reference videos and 3 audio tracks per request.

How do I tell the model which reference to use? Cite each item from metadata.content[] in your prompt as @image1 to @imageN.

Am I charged if a generation fails? No. Failed or errored requests are never billed.