Wan 3.0 inputs: text, image, video, audio on Alibaba vs Sume wan-3.0

Alibaba's Wan3.0-Video page documents text, image, video and audio inputs, with first-frame and first-and-last-frame image-to-video, reference-based generation, up to 30 seconds, at 480P, 720P and 1080P. Sume's wan-3.0 covers text-to-video, image-to-video with an end frame, and reference-to-video, at 2 to 30 seconds and the same three resolutions.

Input modes side by side

The Alibaba column is from the Model Studio doc, last updated September 28, 2026. The Sume column is from the Video Router and Video Generation docs.

How Sume picks the mode

The Sume Video docs say Wan 3.0 accepts audio and video references, along with the Seedance 2.x and MiniMax H3 models. Gemini Omni Flash 1.1 and h3-max-recast do not take audio references.

Mode is inferred, not declared. If frame_images is present, Sume runs image-to-video. If input_references is present, it runs reference-to-video. With neither, it runs text-to-video. If you send both, frame_images wins.

A first and last frame request

A first-and-last-frame request to wan-3.0 on Sume looks like this. At 720p, 10 seconds bills 10 x $0.125 = $1.25 (a $0.10 fal list rate times 1.25).

{
  "model": "wan-3.0",
  "prompt": "The door opens and morning light fills the room",
  "duration": 10,
  "resolution": "720p",
  "frame_images": [
    {"type": "image_url", "image_url": {"url": "https://example.com/first.png"}, "frame_type": "first_frame"},
    {"type": "image_url", "image_url": {"url": "https://example.com/last.png"}, "frame_type": "last_frame"}
  ]
}

Limits of this comparison

The vendor column comes from Alibaba's page, which does not describe any other input type, so this post makes no claim about one. The Sume side is limited to what the Sume docs list. Read supported_input_references and supported_frame_images from GET /v1/videos/models for the live answer, because a model that does not advertise a reference type gets a 400.