Which video model for a 12 s 9:16 clip with end frame and sound?

All four models take a first and last frame and a 12-second duration at 9:16 on Sume. MiniMax H3 at 768p is the cheapest at $0.90, Wan 3.0 at 720p is $1.50, Kling 3 with sound is $2.52, and Seedance 2.5 at 720p is $6.94. Sound is the differentiator: it is a toggle on Kling 3, native on H3, and listed as audio on Seedance 2.5 and Wan 3.0.

Fit check, as listed on Sume

The cluster covers four catalog ids. The first thing to check is whether each accepts 9:16, 12 seconds and an end frame; the second is what sound costs.

How each vendor frames it

Alibaba's Model Studio page, read 2026-10-08, lists 480p, 720p and 1080p for Wan 3.0 up to 30 seconds. MiniMax's H3 post describes native stereo sound with no separation of voice, effects and music, so you cannot mute dialogue only. fal's Kling v3 Pro page lists $0.112 per second silent and $0.168 with audio. fal's Seedance 2.5 explainer lists up to 30 seconds and token billing.

The pick

Choose on what you need from the end frame and the audio.

• Cheapest end-frame clip with audio: Wan 3.0 at $1.50, or $0.75 at 480p.
• You want dialogue or music baked in and one price knob: Kling 3 with sound, $2.52.
• You want to feed an audio track as a reference: Seedance 2.5 or Wan 3.0.
• You want a native stereo bed that you cannot toggle: H3, and expect to mute it in post if you do not want it.

Submit it and read the result

Sume reserves the billable amount from your workspace USD balance when the job is submitted, then returns a polling_url. For the other three models change model and check the GET /v1/videos/models entry for the resolutions it lists. Poll the job until status is completed or failed; a completed job lists its files in unsigned_urls.

import os
import time
import requests

H = {"Authorization": f"Bearer {os.environ['SUME_API_KEY']}"}
body = {
    "model": "wan-3.0",
    "prompt": "A slow push-in on the scene, soft natural light",
    "duration": 12,
    "resolution": "720p",
    "aspect_ratio": "9:16",
}
r = requests.post("https://api.sume.com/v1/videos", headers=H, json=body, timeout=60)
r.raise_for_status()
url = r.json()["polling_url"]
while True:
    job = requests.get(url, headers=H, timeout=30).json()
    if job["status"] in ("completed", "failed"):
        break
    time.sleep(10)
print(job["status"], job.get("unsigned_urls"))

Not covered

Kling 4.0 is not in the Sume catalog, so it is not in this table. Wan 3.0's default is 1080p if you omit the resolution, which would change the Wan figure above; pin it.