Ever wondered how a viral video was actually made? rea-video answers that shot by shot. It's a free, open source agent skill with over 5K stars on GitHub that tells you whether each shot was filmed, generated with AI, rendered in 3D or built in code, then hands you a plan to make your own version.

This kit gets you set up in about 5 minutes and gives you the prompts we use to go from breakdown to remake.

Left: TikTok ad filmed on a phone. Right: an all AI remake.

Left: a TikTok ad filmed on a phone. Right: an all AI remake. rea-video's verdict: the left is all live footage, the right is all AI, 8 shots across 3 generations.

What you need

  • Claude Code, Codex or any AI agent that can run commands

  • ffmpeg and Python 3.10 or newer (your agent can install them for you)

  • The video you want to break down, saved on your computer

Step 1: Install it with one message

Open your agent and send this, with the path to your video at the end:

Install https://github.com/JuneYaooo/rea-video, then break down this video: /path/to/video.mp4

Your agent installs the tool, learns how it works and runs the full breakdown. If it asks to run commands, allow it. That's how it installs.

Tip: not sure what the path is? Drag the video file into your terminal window and it pastes the path for you.

Step 2: Read what you get back

  • A one line verdict, for example "41% AI video + 59% code UI, 8 s of real screen recording, library music at 122 BPM"

  • A remake plan split into what to film, what to generate and what to code

  • report.html: open it in your browser to see a still of every shot, the evidence behind each call, camera and lighting notes, captions, cuts and the audio breakdown

Ask for the report in English and it writes it in English.

Step 3 (optional): Make it sharper with a vision model

Give your agent the endpoint and key for a model that can watch video, such as Gemini. The calls get more accurate and you get a ready to use prompt for every AI shot. This sends frames of your video to that provider, so use a key with a spending limit.

It works fine without one. The one case where it matters: videos saved from TikTok or Instagram are recompressed, and without a vision check accuracy on those drops to about 89% in the repo's own tests.

4 prompts to go from breakdown to remake

1. Find out how a viral video was made

Break down this video with rea-video and explain in plain words how it was made: what was filmed, what was AI and what was built in code. [video path]

2. Get a shot list you can film with your phone

Using the rea-video report for this video, write a shot list the user can film with just a phone: camera position, framing, lighting and what happens in each shot.

3. Remake the AI shots with your own subject

From the rea-video report, take every AI shot and rewrite its prompt so it shows [your product or character] instead. Keep the same camera move, timing and style.

4. Copy the pacing, not the footage

From the rea-video report, give the editing recipe: cut timing, caption style and when captions appear, music BPM and where each sound effect lands, so the same pacing can be rebuilt in CapCut.

Use it the right way

rea-video is for learning how a video is built. Remake it with footage, music and characters you have the rights to, and label AI content wherever your platform asks you to.

You're reading The AI Leverage, where we help you get sharper with AI every week. Tools, workflows and kits like this one, always free.

by bestapps.ai on instagram