AI video for online courses - practical B-roll

A course cutaway should clarify the narration. It does not need to impersonate the instructor or contain the whole lesson. Start with one object and one action, such as coffee brewing in a glass vessel. Add instructions and step numbers in your editor so you can verify every word.

Prepare the shot

• Mark the exact narration sentence and the intended cut-in and cut-out points.
• Choose an object instead of a person when facial expression is irrelevant to the explanation.
• Request a static camera or one slow movement. Avoid combining a pan, orbit and zoom in one shot.
• Review a single test without sound. Inspect object shape, hands, packaging text and continuity.
• Place the clip under the finished voiceover. Keep usable still moments at each end for editing.
• Watch an exported lesson on a phone. The illustration must leave room for captions.

Review before publication

• The image matches the narration.
• The object keeps its shape throughout the shot.
• Lesson captions remain readable at phone size.

API input example

The JSON below is the request body for POST /v1/generate. Use your own accessible image URL where a reference is required.

{
  "model": "kling-3.0",
  "input": {
    "prompt": "Close-up of coffee slowly dripping into a glass vessel, static camera, soft daylight, neutral background, no text",
    "duration": 5
  }
}

Model and cost

Use Kling 3.0. Its current unit price is from $0.169 per second. Check the pricing page and calculator for your settings. Credits never expire, there is no subscription and failed generations are free. A successful output is billable even if you choose not to publish it.

FAQ

Should the model generate instructional text inside the shot?

Add instructional text in your editor. You can correct a word without generating the clip again.

Can I check the request before a paid run?

Yes. Check the model schema and the cost shown in Studio. Local demo API tests validate request handling but do not prove the visual or audio quality of the real model.

Where are the supported parameters?

Model details and the generation API guide list the supported inputs. Preview mode does not provide a real generated result.