First and last frame video on Sume: Seedance 2.5 or Wan 3.0

Send frame_images with a first_frame and a last_frame entry on POST /v1/videos, and either seedance-2.5 or wan-3.0 will animate between them. The Video Router catalog lists both for start/end-frame image-to-video.

If you also send reference inputs, frame_images wins and the request is treated as image-to-video.

Request

{
  "model": "seedance-2.5",
  "prompt": "The door opens and light floods the room",
  "duration": 10,
  "resolution": "720p",
  "frame_images": [
    {"type": "image_url", "image_url": {"url": "https://example.com/a.png"}, "frame_type": "first_frame"},
    {"type": "image_url", "image_url": {"url": "https://example.com/b.png"}, "frame_type": "last_frame"}
  ]
}

10 s cost

Tips

• Match aspect ratio to the images.
• Make the two frames plausibly connected.
• Use HTTPS image URLs.

Planning the two frames

The best start/end-frame clips connect two states of one scene: a door closed then open, a product boxed then unboxed. Give the prompt the action that links them.

If the frames are very different, the model will invent the bridge, so keep camera angle and subject consistent.

Frames versus references

Frame images set the first or last picture. References guide style and content without fixing a frame. The docs state that sending both makes frame_images control the mode, so pick one mode per job.

• Use first frame only for a simple animate-a-photo job.
• Add a last frame for a controlled ending.
• Use references for identity across scenes.