Classroom science explainer clips: six 10-second AI shots, priced

To illustrate a science lesson with AI video, write one prompt for each idea and render six 10-second clips. On Sume the set costs $6.00 on MiniMax H3 Max at 768p, $7.50 on Wan 3.0 or Gemini Omni Flash 1.1 at 720p, and $34.68 on Seedance 2.5.

AI video is an illustration tool and not a lab. The model draws what looks plausible, so a clip of a volcano or a cell can show the wrong structure. Use it for mood and context and keep diagrams and numbers in your own slides.

The cost of a six-shot lesson

Prices are the vendor list rate times 1.25, rounded up per clip, for a 10-second 16:9 text-to-video request. Seedance is billed per token at 1280x720.

What to ask the model for

• Context shots: a rainforest canopy, a night sky, waves on a beach, steam over a kettle.
• Slow, continuous motion: a time-lapse of clouds, a camera moving across a landscape.
• Abstract motion for a title: particles, light, fluid.
• Do not ask for labelled diagrams, equations, or the inside of an organism. The model will invent them.

Write each prompt for one idea

A good prompt for a class names one subject, one camera move and the light: "slow aerial pass over a volcanic crater at dawn, steam rising, no people". Avoid chaining ideas, such as "a volcano erupts and then a city is covered in ash", because the model may skip or blend the steps.

Render a 480p test first at $0.63 on Wan 3.0, and only raise the resolution once the shot is right. This request uses Wan 3.0 at 480p for the test.

import os
import time

import requests

API = "https://api.sume.com/v1/videos"
HEADERS = {"Authorization": "Bearer " + os.environ["SUME_API_KEY"]}
body = {
    "model": "wan-3.0",
    "prompt": "Slow aerial pass over a volcanic crater at dawn, steam rising, no people, natural colour",
    "duration": 10,
    "resolution": "480p",
    "aspect_ratio": "16:9",
}
job = requests.post(API, headers=HEADERS, json=body, timeout=60)
job.raise_for_status()
poll_url = job.json()["polling_url"]
while True:
    state = requests.get(poll_url, headers=HEADERS, timeout=60).json()
    if state["status"] in ("completed", "failed", "cancelled"):
        break
    time.sleep(10)
print(state["status"], state.get("usage"), state.get("unsigned_urls"))

Check before it reaches students

Watch every clip for facts you did not ask for: a wrong number of legs on an insect, a river that flows uphill, a crater with the wrong shape. If a clip is wrong in a way that matters, replace it with a photo or a diagram from a source you trust. Say in the lesson that the footage is AI-generated, so students treat it as an illustration and not as evidence.

For the budget, six shots at 720p on Wan 3.0 cost $7.50 for the first pass. Add a second pass for the two or three shots you redo and a reserve of about $4 for the lesson.

If the lesson needs one longer sequence instead of six short ones, Wan 3.0 and Seedance 2.5 accept up to 30 seconds in one job. A 30-second Wan 3.0 clip at 720p costs $3.75, which is half the price of the six-shot set. The trade-off is that a long clip gives the model more time to drift away from the idea in your prompt, so a set of short shots is easier to check and to replace one by one.