Wan 3.0 API: 2–30 second AI video with audio, frames, and references

Wan 3.0 is available through the Sume API as wan-3.0 on POST /v1/videos, for clips of 2 to 30 seconds at 480p, 720p, or 1080p. It takes a prompt alone, a first and optional last frame, or image, video, and audio references, and it can generate an audio track with the clip.

The Video generation docs give Wan 3.0's duration range and reference types; the rest below comes from the model catalog the API serves, read on 2026-09-26. Submitting a job and fetching the result work as for any catalog model, as An OpenRouter-compatible video API shows.

What can Wan 3.0 do on Sume?

These values come from the wan-3.0 catalog entry that GET /v1/videos/models reports.

How long can a Wan 3.0 clip be?

From 2 to 30 seconds. duration is an integer, and the descriptor's supported_durations lists every whole second from 2 through 30. Any other positive integer is refused at submit with 400 unsupported_capability, and the error's details.supported carries the accepted list. For a video longer than one clip, see AI video length limits by model.

How do I start a clip from a first and last frame?

Send frame_images with a first_frame and, if you want a fixed ending, a last_frame. A last_frame without a first_frame is refused. Image to video with first and last frames covers frames across models.

curl -X POST "https://api.sume.com/v1/videos" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: wan-frames-001" \
  -d '{
    "model": "wan-3.0",
    "prompt": "A hiker walks from the forest edge to the lake shore at sunset",
    "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" }
    ],
    "resolution": "1080p",
    "duration": 12
  }'

How many references can Wan 3.0 take?

Reference-to-video uses input_references. Wan 3.0 honors audio and video references as well as images, within these limits:

• Up to 10 image references.
• Up to 5 video references, 15 seconds in total, each at 16 fps or more.
• Up to 5 audio references, 15 seconds in total.

Does Wan 3.0 generate audio?

It can, and the catalog describes that audio as optional. generate_audio defaults to the model's audio capability, which is on for Wan 3.0, so Sume asks for an audio track unless you send generate_audio: false.

What does Wan 3.0 not support on Sume?

The catalog and the /v1/videos rules list these gaps:

• Editing an existing clip.
• Aspect ratios outside 16:9, 4:3, 1:1, 3:4, and 9:16.
• bitrate_mode.
• seed and size: no v1 model accepts them, and each returns 400 unsupported_parameter.
• file_url / web_url and enable_thinking: not exposed in v1.

How is Wan 3.0 billed?

Per output second, at a rate set by resolution: pricing_skus holds one per-video-second-<resolution> rate each for 480p, 720p, and 1080p. List video generation models via API explains what those SKUs mean and where the 5.5% default agent fee comes in.