A first-frame still plus a 30-second Wan 3.0 clip: the pair's cost

Making the first frame yourself adds between $0.005 and $0.10 to a 30-second Wan 3.0 clip, depending on the image model: Higgsfield Soul is $0.005, Seedream 5 Lite $0.0437 and Nano Banana 2.1 $0.10 per image on Sume. A 720p clip is $3.75, so the still is a rounding error on the video, but it lets you approve the look before you pay for motion.

Total cost of the pair

One still plus one 30-second clip, Sume list price:

Why the still is worth its price

Wan 3.0 supports first and last frame input; Alibaba lists first/last frame for wan3.0-video (read 2026-10-08), and the frame_images field in the Video generation docs carries a frame_type of first_frame or last_frame. Starting from an approved image fixes the subject, framing and colors, so you are judging motion rather than gambling on the opening shot.

Try three stills, pick one, then spend on video. Three Soul stills are $0.015, less than a third of one Nano Banana 2.1 still.

The video request

The image URL must be public HTTPS.

curl -X POST https://api.sume.com/v1/videos \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: still-to-30s-001" \
  -d '{
    "model": "wan-3.0",
    "prompt": "The camera pulls back slowly as steam rises from the cup",
    "duration": 30,
    "resolution": "480p",
    "frame_images": [{
      "type": "image_url",
      "image_url": { "url": "https://example.com/first-frame.png" },
      "frame_type": "first_frame"
    }]
  }'

Making three tries on the still

The cheapest way to get a good opening frame is volume at the low end: several Soul or Seedream 5 Lite stills cost less than one extra video attempt. Choose the frame, check that the subject and framing are what you want, and only then submit the video. The image URL has to be reachable over public HTTPS when the video job starts.

A sensible order

Approve the still, render the clip at 480p, and move to 1080p only for the take you keep. The 480p pass costs a quarter of 1080p and shows whether the motion works.