DEV Community

Xinyue Zhan
Xinyue Zhan

Posted on

Reverse-engineering a reference video: a Rea Video walkthrough

I had a competitor clip that worked and no idea how to reproduce the look. This is the process I used, in order, with the tool that did each step.

Step 0: describe the reference before touching a tool

Watch the clip twice on mute. Write down, in this order:

  1. Shot count and cut rhythm — one shot or five? Cuts on beat or on motion?
  2. Framing per shot — wide, medium, close. Where is the subject in frame?
  3. Camera behaviour — static, push-in, drift, handheld. This is the one people skip, and it is the one that decides which tool to use.
  4. Light — direction, hardness, colour temperature.

If you cannot describe the shot, no tool will reproduce it. The description is the input.

Step 1: re-shoot the shot with Rea Video

Rea Video is reference-driven: hand it the clip, describe the new subject, and it re-creates the shot with your subject in it. Two things I learned:

  • Feed a reference where the camera work is legible. Fast cutting confuses the recreation; a clean 3-second push-in does not.
  • In the description, change only the subject. If you also describe staging, you are asking it to invent, and it will drift.

The output is a first pass. Expect to re-roll two or three times per shot.

Step 2: fix the single-shot gaps with Wan 3.0

Some shots in the reference are one clean beat — a push-in on hands, a slow reveal. For those, re-shooting is slower than generating. Wan 3.0 does text-to-video and image-to-video with motion presets, and the preset is the useful control: pick slow push-in and the clip reads right before you touch the prompt.

Where I used it instead of Rea Video:

  • Shots with no subject to swap (backgrounds, transitions, cutaways).
  • Shots where the reference was too low-res to re-create cleanly.

Step 3: assemble, caption and dub in Pavo AI

Once the shots exist, you have a sequence, and a sequence needs a timeline. Pavo AI covers generation, timeline editing, captions and multilingual dubbing in one workspace. The workflow that finally worked:

  1. Drop all shots on one timeline, 9:16 master.
  2. Cut to the reference rhythm — this is where you find out you generated two shots too many.
  3. Captions next, because they change where the eye lands.
  4. Dub last. Dubbing before the cut is locked means re-dubbing every time you trim a shot.
  5. Export 16:9 from the same timeline for the landing page.

Character consistency is the part that makes this viable: lock the reference frame once, reuse it for every shot, and shot three still matches shot one. Without that lock I was re-generating whole sequences.

Prompt examples for the locked-reference approach: pavoai.app/prompts.

What the split actually looks like

Step Tool Why this one
Re-create a shot from a reference Rea Video it re-shoots instead of generating from scratch
Generate a missing single shot Wan 3.0 motion presets, no timeline overhead
Cut, caption, dub the sequence Pavo AI generation and edit share one tab

The honest summary

Most of the work is step 0. The tools are fast; the description is slow. If you skip the shot description and start generating, you will burn an hour re-rolling and end up writing the description anyway.

Links: reavideo.pro · wan-3.io · pavoai.app

Top comments (0)