YouTube video to shot-by-shot AI prompts

A coffee ad can look simple until you try to describe it. Does the camera move towards the cup, or does someone lift the cup towards the camera? That one detail changes the prompt.

To turn a YouTube reference into useful video prompts, write down what happens in each shot before you worry about style. Start with a short clip, check its cuts, then write one prompt per shot. A few accurate sentences beat a paragraph full of “cinematic” and “beautiful.”

Open the right input tab

In NanoPhoto.AI's video-to-prompt tool, use the YouTube tab for a watch-page link. Use the upload tab for a video file you already have permission to use. The screenshot below shows the live YouTube input, captured on September 10, 2026.

Check the limits beside the input before submitting. A YouTube page, a direct MP4 link and a file on your computer are different things; a tool that accepts one may not accept the others. If a link fails, check its public availability before trying an authorized local copy.

For a first attempt, pick a section with a clear subject and a handful of cuts. A long compilation makes it harder to tell whether the analysis missed something or whether the reference itself keeps changing style.

Put the cuts on paper

A scene can contain several shots. In the coffee example, the kitchen is one scene; the wide view of the table and the close-up of the pour are two shots.

Watch once for the action, then again for the cuts. Check quick inserts and dissolves on the timeline. Treat AI timestamps as estimates until you have compared them with the source. PySceneDetect is useful when you need a separate cut-detection step; it locates changes rather than explaining what the person is doing.

Use this table as your own worksheet:

The table is a worksheet you can copy. NanoPhoto's output may use a different layout.

Try the three-shot coffee example

The cover illustration was generated for this guide. Here is the fictional sequence behind it: establish the cup, pour the coffee, lift the cup. The times below are a plan, not measurements from a processed video.

For the middle shot:

Close-up of coffee pouring from a small steel pot into a matte blue
ceramic cup on a pale wooden table. Soft morning light comes from
the left. The camera stays still as the coffee level rises.
A thin wisp of steam drifts upward. Warm natural colors, soft background.
Keep the cup's shape and blue glaze unchanged.

Now remove “the camera stays still.” The generator has to make that decision itself. This is why concrete observations matter more than extra adjectives.

Set duration and aspect ratio in the generator's controls when available. Give the next cut its own prompt so you can change it without rewriting the whole sequence.

Keep the cup blue after the cut

Write a shared note before generating: matte blue cup, pale wooden table, light from the left. Reuse those details across the shots. If the generator supports a reference image, use one to help hold the cup's appearance; repeating its name alone does not guarantee a match.

Review the first shot before making the rest. Look for the requested action and camera movement. If the cup changes shape, fix that problem first. Adding “professional, beautiful, cinematic” will not tell the model which shape you meant.

Keep two sets of notes: what you observed in the reference, and what you decided to change. That makes revisions easier when someone asks why the final scene differs from the source.

Can I recover the creator's original prompt?

The video does not contain all the original instructions or settings. Reverse prompting gives you a description to work from, not access to those hidden choices.

Should I remove subtitles first?

Only when they obstruct something you need to inspect or reuse. Keep the original. If you need a clean copy for editing, see how to check subtitle removal without cropping.

For recurring work, use the manual-versus-automatic comparison to measure how much checking and rewriting each approach leaves you.

NanoPhoto.AI makes the tool used in this guide. We checked the public interface on September 10, 2026. Writing and the cover illustration used AI assistance; the coffee sequence is a teaching example.