Wan 3.0 on Sume: one id, three modes, and which field picks each

You send model: "wan-3.0" and the fields you have; Sume picks the mode. No frame_images and no input_references means text-to-video, a frame_images entry means image-to-video, and input_references alone means reference-to-video. If you send both, frame_images wins.

fal's Wan 3 page, read 2026-10-05, lists three endpoints (text-to-video, image-to-video, reference-to-video) at one rate card: $0.05, $0.10 and $0.20 per second at 480p, 720p and 1080p. Sume bills the same tiers at list times 1.25.

Field to mode

The Sume side of this is documented in the video generation docs: the API infers the mode from the body, and you never declare it.

A first-frame request

The body below pins the opening frame. A single reference image that should condition the whole clip belongs in input_references instead.

{
  "model": "wan-3.0",
  "prompt": "Slow push-in on the mug, steam rising",
  "duration": 6,
  "resolution": "720p",
  "aspect_ratio": "16:9",
  "frame_images": [
    {
      "type": "image_url",
      "image_url": { "url": "https://example.com/mug.png" },
      "frame_type": "first_frame"
    }
  ]
}

What stays the same across modes

The per-second price does not change with the mode on either side, so a 6-second 720p clip is $0.75 whether it starts from text, a first frame or references. Wan 3.0 on Sume takes 2 to 30 seconds at 480p, 720p and 1080p, so the whole range of modes is available at the same length limits.

One id also means one place to look when something fails. If an input URL cannot be fetched, the poll response says so with the same public error text, whichever mode you used.