45-second AI video API: one 30 s job plus one 15 s job on Sume

A 45-second AI video on Sume takes two jobs, because the longest single clip is 30 seconds. The practical split is 30 + 15. Wan 3.0 covers both halves at $0.0625, $0.125 or $0.25 per second billed at 480p, 720p and 1080p, so the whole 45 seconds costs $2.82, $5.63 or $11.25.

Seedance 2.5 can do the 30-second half too, but at about $17.34 for 30 seconds at 720p it is the premium route.

What does 45 seconds cost on each route?

Sume bills provider list times 1.25 and rounds each job up to the cent. The 30-second and 15-second jobs are priced separately and added.

Why not stay on one model for both halves?

Staying on one model keeps the look and the audio character close at the join. Mixing a Wan 30-second job with a Kling 3 15-second job is possible, since Kling also reaches 15 seconds, but Kling takes no reference images, so it cannot hold a character from the first clip.

If the two halves need to share a subject, pin the same model and pass the same input_references to both requests.

How should you place the cut?

Plan the cut at a natural break, such as a scene change, instead of in the middle of a move. Write the 30-second brief so it ends on a still beat, and start the second job from the last frame of the first with frame_images and a first_frame. A fixed starting frame is the most reliable way to keep the set and the lighting the same.

• Job A: 30 seconds, ends on a held shot.
• Job B: 15 seconds, first_frame taken from the final frame of job A.
• Join the two files in your editor or timeline; Sume does not return one merged file.

What does the request look like?

Each half is its own POST /v1/videos. Keys come from the environment, and idempotency keys keep a retry from billing twice.

import os, requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}

for key, seconds in (("ad-45-a", 30), ("ad-45-b", 15)):
    r = requests.post(
        "https://api.sume.com/v1/videos",
        headers={**H, "Idempotency-Key": key},
        json={
            "model": "wan-3.0",
            "prompt": "A barista makes a latte, warm morning light",
            "duration": seconds,
            "resolution": "720p",
            "aspect_ratio": "16:9",
        },
    )
    r.raise_for_status()
    print(key, r.json()["id"])

Is a shorter plan cheaper?

Yes. Draft the whole thing at 480p first, which costs $2.82, then regenerate only the clips you keep at 720p or 1080p. Check usage.cost on the first completed job to confirm the billed amount before you scale the batch.