How to Turn a Reference Video Into Shot-by-Shot AI Video Prompts

To turn a reference video into AI video prompts, split it at every cut, log each shot's time range, and describe five elements per shot: shot size, camera move, duration, action and light. Each row then becomes one prompt for a video model on MakerNice, such as ByteDance's Seedance 2.5, Alibaba's Wan 3.0, Kuaishou's Kling 3.0 Pro or MiniMax H3.

How do you break a reference video into shots?

Watch the clip at reduced speed, or step through it frame by frame, and mark every cut. Write down where each shot starts and ends; that list of time ranges is your shot list. You can also ask an AI assistant that understands video to list the shots with timecodes first, then check every boundary by hand, because quick cuts are easy to miss.

• Start a new row at every hard cut, not at every change in what the subject does.
• Keep one row for a shot whose camera keeps moving, and note where the move changes.
• Round each shot to whole seconds, since the duration field takes integers.
• Note which shots feel fast and which linger; that rhythm is what you are recreating.

What five elements should you record for each shot?

How do you turn the five elements into one prompt?

Write the elements in a fixed order: shot size and type, your subject, one action, one camera move, then light and mood, and finish with a pace that fits the length. Put the length itself in the duration field rather than in the prompt. Swap the reference's people, products and settings for your own, and keep one action and one camera move per prompt. The template below is plain text you can copy into a notes file or spreadsheet.

SHOT LIST (one row per cut)
#  | time      | shot size / type | camera move   | action                       | light / mood
01 | 0:00-0:04 | extreme close-up | static        | steam rises from the coffee  | warm low morning light
02 | 0:04-0:09 | medium shot      | slow dolly-in | a hand slides the mug forward | window light from the left

PROMPT TEMPLATE (one prompt per shot)
[shot size and type] of [your own subject], [one action], [one camera move], [light and mood], [pace that fits the duration].
duration: [whole seconds within the model's range]

EXAMPLE (shot 02)
Medium shot at table height of a white ceramic mug on a wooden counter, a hand slides the mug toward the camera, slow dolly-in, warm window light from the left, calm and steady pace.
duration: 5

What does a finished shot list look like?

This example logs a 17-second clip of your own mug in four shots of 4, 5, 4 and 4 seconds. Shot 02 is the row filled into the template above; the other three follow the same pattern, so the list becomes four prompts with four duration values.

How long can each shot be on each model?

The duration field takes whole seconds, and each model accepts its own range; leave it out and you get the 5 second default. Reference videos often have cuts shorter than a model's minimum. In that case, generate at the minimum and trim the clip in your editor, or pick a model whose range covers the shot.

How do you generate a whole shot list in batch?

Treat every row as one generation. Keep the model, resolution and aspect_ratio the same for every shot so the clips match, repeat your subject description word for word, and draft at a low resolution before the final render. Through the API, loop over the rows and quote and create one generation per shot, each with its own Idempotency-Key. For tighter framing, use image-to-video and pass a first frame you made in image_url. MakerNice returns each shot as a separate clip; cutting them together in order happens in any video editor you like.

What should you not copy from a reference video?

Work only from footage you shot yourself or have the right to use. Recreate the structure: the pacing, the order of shots and the framing. Do not reproduce someone else's video, characters, logos or likeness. Use your own product or scene, and AI-generated characters for people; even with consent, the Acceptable Use Policy does not allow placing real people in scenes they never took part in.

For wording each element, read the text-to-video prompt tips post. Then open a model page and try your first shot in the Playground, or check the Pricing page before you render the whole list.