Language school promo, 15 s at 9:16: Seedance 2.5, Wan or Kling

For a 15-second vertical teaser, start on Wan 3.0 at 720p: $1.88 on Sume. Move to Kling 3 with sound ($3.15) if you want the cleaner motion of a 1080p image-to-video start, and to Seedance 2.5 ($8.67 at 720p) when you want to feed it a voice or music reference. All three accept 9:16.

What a language-school teaser needs

The usual teaser is a teacher at a whiteboard, a student speaking, and a text card with a class time. The text card should be added after generation, since Alibaba's post says on-screen text accuracy is still improving. What you ask the model for is a believable classroom and a spoken moment.

Seedance 2.5's reference-to-video takes audio files as well as images and video. fal's explainer, read 2026-10-08, lists up to 50 reference files, of which at most 10 are audio. Wan 3.0 on Sume lists up to 5 audio references. Kling 3 on Sume lists no reference inputs at all, so a voice you want to match can only come in through the prompt.

Prices for 15 seconds, 9:16

Sume bills provider list times 1.25, rounded up to the cent per job.

How to pick

Pick on the thing that is hard to fix afterwards.

• Cheapest believable draft: Wan 3.0 at 480p, $0.94.
• Best chance of a controllable look from a still: Kling 3 with a first frame, 15 seconds at $2.10 silent.
• Matching a specific voice or backing track: Seedance 2.5 with an audio reference, and draft at 480p ($4.04) before the 720p final.
• Several languages: generate the visual once, silent, and add each language as a voiceover and caption pass instead of regenerating the clip per language.

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. The same body works for the other two models; change model and drop resolution values the model does not list. 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": 15,
    "resolution": "480p",
    "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"))

Limits

Seedance 2.5 and Wan 3.0 both go to 30 seconds on Sume, Kling 3 to 15, so a longer teaser changes the choice. I did not test any of the three on pronunciation or lip accuracy in a specific language, and the vendor pages I read do not make per-language claims for Seedance 2.5 or Kling 3. Test a draft with your target language before a campaign.