Image to videoCreative and entertainment

New scenes from a reference clip

A reference clip's motion and framing are reused for a brand new scene. For editors and creative teams.

$0.10 a secondwith Wan 3.0 Reference-to-Video. Three models to choose from below.

You senda reference video and a prompt for the new versionYou getvideo

Try it in the playground See the code

How it works

Send a reference video and a prompt for the new version
One POST to /run with the model id and your input. It answers at once with an id.
We run Wan 3.0 Reference-to-Video
$0.10 a second. Charged when it starts, refunded by itself if it fails.
Collect a video
GET /generations/{id}?wait=55 returns a new clip that follows the reference, as a file URL. Or pass a webhook and we call you.

The model for this

Prices are read from the live catalog. Each one opens the playground with this use case filled in.

Balanced
Wan 3.0 Reference-to-Video
$0.10 a secondThe one most people should start withwan-3-r2v

The request

Filled in with sample files from our CDN, so it runs as it is. Swap in your own URLs.

POST /run

curl https://api.percify.io/v3/playground/v1/run \
  -H "Authorization: Bearer $PERCIFY_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "modelId": "wan-3-r2v",
    "input": {
      "prompt": "The same camera move, but on a snowy mountain ridge at dawn",
      "reference_videos": [
        "https://cdn.percify.io/media-assets/playground-examples/infinitetalk-fast/4.mp4"
      ]
    }
  }'

Full reference in the docs. Every field Wan 3.0 Reference-to-Video takes is on its model page.

Questions

What do I send to new scenes from a reference clip?

Send a reference video and a prompt for the new version in one POST to https://api.percify.io/v3/playground/v1/run. You get a new clip that follows the reference.

What does it cost?

With Wan 3.0 Reference-to-Video it is $0.10 a second (5 credits a second at the lowest setting; higher resolutions and audio cost more). The estimate call returns the exact price of your input before you run it, and a run that fails is refunded.

Which model should I use?

Balanced: Wan 3.0 Reference-to-Video ($0.10 a second). Switching is one string in the request: the model id.

Can an AI agent do this for me?

Yes. Connect your agent to the MCP server at https://mcp.percify.io/mcp and ask for it in plain words. The agent calls the generate tool with the same model id and input shown on this page, and the call spends the same credits.

Related use cases

How it works

Send a reference video and a prompt for the new version

One POST to /run with the model id and your input. It answers at once with an id.

We run Wan 3.0 Reference-to-Video

$0.10 a second. Charged when it starts, refunded by itself if it fails.

Collect a video

GET /generations/{id}?wait=55 returns a new clip that follows the reference, as a file URL. Or pass a webhook and we call you.