Skip to main content
GUIDE

Generate video between first and last frames

First/last-frame mode uses two images to constrain the opening and ending frames. The model detects the mode from references, which must contain exactly one first_frame and one last_frame.

Last updated

Prepare the frames

  1. Host two imagesUpload both frames to publicly reachable HTTPS URLs.
  2. Reference them in orderUse @图片1 and @图片2 in the same order as references.
  3. Assign the rolesThe first item is first_frame and the second is last_frame.
  4. Submit and pollWait for completed, then read metadata.url.
First and last frame requestjson
{
  "model": "seedance2.0",
  "prompt": "The person moves naturally from @图片1 to @图片2 while the camera stays stable",
  "seconds": 5,
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "references": [
    {
      "type": "image",
      "role": "first_frame",
      "source": "https://cdn.example.com/first-frame.png"
    },
    {
      "type": "image",
      "role": "last_frame",
      "source": "https://cdn.example.com/last-frame.png"
    }
  ]
}

Common causes of failure

  • Only one of the two frames is submitted.
  • A first or last frame is not an image.
  • First/last frames are mixed with reference video or audio.
  • The prompt omits a submitted material or uses a different order.
  • A localhost, private IP, HTTP, hotlink-protected, or signed temporary URL is used.
Reference to video
Combine image, video, and audio references.
Seedance 2.5
Review the higher reference limits.