Production guides

How to reverse engineer a video, shot by shot

Video reverse engineering asks how a reference was produced, not just what happens in it. The useful output is a plan: which shots to film, which assets to generate, which interfaces to rebuild and which decisions still need evidence.

Start with the result you want to learn

Choose a concrete goal such as product framing, dialogue pacing or interface animation. Watch the full reference and record its structure before inspecting individual shots. A transcript helps with speech, but it cannot describe camera movement, compositing or the source of the picture.

Classify layers instead of labeling the whole video

Inspect live footage, generated pictures, 3D renders, code graphics, recordings and still images separately. Then record captions, stickers and picture-in-picture. A mixed video can use several of these in one shot. Treat the classification as a working hypothesis, not proof of the original software or model.

Map edits and sound to visible events

Measure shot lengths and note hard cuts, overlaps and changes of speed. Listen for speech, background music and short sound effects separately. Link an effect to the action it emphasizes, such as a product landing on a table. When you change the timing, move the effect with the action instead of copying its old timestamp.

Build and test one representative shot

Prepare the assets you own or can use, then test the most important shot. For an interface demo, rebuild text and layout in code and provide your own screen recording. For a live product shot, reproduce framing and lighting before adding effects. Review composition, timing and readability before extending the workflow to the rest of the video.

What the analysis cannot recover

Compressed video rarely reveals an exact original prompt, editable project or music license. Mark these gaps explicitly. A recreation plan is not a finished video, a guarantee of identical output or permission to reuse someone else’s material. rea-video currently runs through an AI coding assistant and local CLI.

A reusable analysis template

Reference and learning goal
Shot timestamps and representative frames
Picture source + overlays + sound
Evidence / alternative explanation / confidence
Required assets and chosen production tools
One-shot test / review notes / next steps

Practice with a real report

The product-promo case combines code-rendered graphics with real recordings. Use the report to identify what can be rebuilt in code and what you need to record yourself; these require different production steps.

Open the case and download its report →

Continue your workflow