Zuverse

Reverse-engineer any video ad. Remake it as your own.

Give Zuverse a TikTok, an Instagram Reel, a Facebook Ad Library ad or any video file. It breaks the ad down shot by shot (the hook, every word, every cut, the person, the place, the camera, the sound), then makes it again with a new person, a new voice and your brand.

It's the same pipeline as the Video Ad Copier in Zuops, open-sourced as a tool you hand to Claude Code. Claude runs it, shows you the plan, asks what you want changed, and tells you the price before anything is spent.

Give it to your Claude

Open Claude Code in any folder and paste:

Set up Zuverse from https://github.com/zutrue/zuverse and remake this ad for my business: <link to the ad>

That's it. Claude clones the repo, installs it, asks for your Zurelay API key, reads the ad, and walks you through the remake.

Or install it as a Claude Code plugin (works in every project)

/plugin marketplace add zutrue/zuverse
/plugin install zuverse@zuverse

Then ask Claude to "remake this ad: ". In plugin mode, install ffmpeg yourself if zuverse doctor says it's missing (winget install Gyan.FFmpeg, brew install ffmpeg or sudo apt install ffmpeg).

What it does

1. Gets the ad. TikTok, Instagram and direct video links download on their own; Facebook Ad Library ads come through your browser (Claude does it for you if it has a browser tool). The first 29 seconds are kept.
2. Reads it shot by shot. Gemini 3.1 Pro watches the clip with its sound and writes the plan: the hook, the transcript, who's on camera, where, how it's filmed, the music, brand mentions and on-screen text. The cuts are measured from the frames, so the remake cuts where the original cuts.
3. Takes your changes. A new script, offer, place, outfit or language: Claude works them into the plan.
4. Casts the new person. GPT Image 2.5 makes a similar but clearly different person (same kind of presenter, never the same face), or you use your own photos.
5. Makes the video. Seedance 2.5 renders it with the new person speaking in a voice that fits them, and the second voice of a phone call kept as a second voice.
6. Edits it, if you want. Off unless you ask: every pause cut out, word-by-word captions that pop, punch-in zooms on the jump cuts, a few stickers and a progress bar. Gemini times the words (a few cents) and Claude picks what to emphasize; ffmpeg does the rest on your computer.

Two ways to remake:

	Replace the person (mode=replace)	Make a new version (mode=new)
What's kept	The original's motion, timing and camera (the clip is the reference)	The shot list: same shots, same cuts, same framing and action
What's new	The person, the voice, the brand	Every person and the place: each shot is redrawn fresh
Best for	The closest copy of a winning ad	A fresh ad nobody can trace back to the original footage
Resolution	480p or 720p	480p, 720p or 1080p
Good to know	Needs a face for the new person (cast or your photo). The video model sometimes declines a new face in real footage; that's free, and a new version is the way around it	Takes a few more minutes: every shot is drawn first

Why Zurelay

Zuverse needs three kinds of AI: a model that watches video (Gemini 3.1 Pro), an image model (GPT Image 2.5) and a video model (Seedance 2.5). Zurelay serves all of them behind one OpenAI-compatible API key, prepaid from $10 with no subscription, at a fraction of each lab's own price. It takes the ad and the photos inline, so there's nothing to host, and failed videos aren't charged. (Full disclosure: Zurelay is made by the same team as Zuverse.)

What	Model on Zurelay	Price (Oct 2026)
Reading the ad	gemini-3.1-pro	about $0.05 an ad
Cast face + shot frames	gpt-image-2.5-sunburst	about $0.03 an image
The video	seedance-2.5	$0.135/s at 480p, $0.305/s at 720p, $0.75/s at 1080p
The original clip as a reference (Replace mode)	seedance-2.5	$0.081/s at 480p, $0.183/s at 720p

What a remake costs:

Video	Make a new version	Replace the person (with a 15 s reference)
15 s at 480p (draft)	about $2.40	about $3.30
15 s at 720p	about $5	about $7.40
30 s at 720p	about $9.70	about $14.50 (29 s reference)

Live prices: zurelay.com/models. zuverse cost shows the price of your exact settings before you render.

Get a key: sign up at zurelay.com, add credit, create an API key, and put it in a .env file next to where you run Zuverse:

ZURELAY_API_KEY=zr_live_your_key_here

Use it without Claude

Zuverse is a plain Node.js CLI (Node 18+, no build step).

git clone https://github.com/zutrue/zuverse && cd zuverse
npm install           # also downloads ffmpeg
cp .env.example .env  # then paste your Zurelay key into .env
node bin/zuverse.mjs doctor

node bin/zuverse.mjs fetch "https://www.tiktok.com/@someone/video/123..." --name roof-ad
node bin/zuverse.mjs read roof-ad
node bin/zuverse.mjs set roof-ad mode=new duration=15 resolution=480p company="Peak Roofing"
node bin/zuverse.mjs photo roof-ad me.jpg --note "the owner, Sam"   # optional: your own face
node bin/zuverse.mjs cost roof-ad
node bin/zuverse.mjs make roof-ad --yes

Or in one go: node bin/zuverse.mjs go "<link>" --mode new --duration 15 --resolution 480p --company "Peak Roofing" --yes

Command	What it does
doctor	Checks Node, ffmpeg, yt-dlp and your key
fetch <link or file>	Gets the ad and trims it to 29 s (--html saved.html --ad <id> for a saved Ad Library page)
read	Reads it shot by shot and measures the cuts
plan	Shows the plan: script, person, place, beats, settings
set key=value ...	Changes settings (below)
photo <file> --note "..."	Uses your own person or product photo (up to 9)
direction-rules	The rules for direction.json (your changes worked into the plan) and the current plan
cast [--again]	Casts a similar but different person
frames [--remap]	New-version mode: maps every shot and draws it fresh
prompt	Builds the Seedance prompt (prompt.txt, editable) and its references
cost	What it will cost
render --yes	Makes the video and downloads it
status	Keeps waiting for a render that's still going
make --yes	cast + frames + prompt + render
go <link> --yes	fetch + read + make
list	Your projects
edit-rules [--video f]	Optional edit: times the words, shows the pauses it would cut and how to write edit.json
edit [--video f] [--auto]	Optional edit: cuts every pause, adds captions, zooms, stickers and a progress bar (zuverse-*-edit-*.mp4)

Setting	Values
mode	replace (default) or new
voice	new (default: the new person's own voice) or original (replace mode only)
duration	4 to 30 seconds (default: the ad's length, up to 29)
resolution	480p, 720p (default), 1080p
aspect	9:16, 16:9, 1:1 (default: the ad's shape)
company	Your business name: every brand mention becomes yours
removeBranding	true strips brand names instead
text	off (default: no captions or on-screen words, which AI video misspells) or on
music	on (default) or off
lookalike	true (default): cast a similar person when you give no photo
images	on (default) or off: send no generated face or shot frames; the video model casts from the words
notes	Your direction, sent with the prompt
script	Your own words for the person to say

Every project is a folder in ./zuverse-projects/ with everything each step made (analysis.json, shots/, frames/, cast/, prompt.txt, the videos), so you can look, edit by hand, and re-run any step.

How it works

link ─► fetch ─► clip.mp4 (29 s, sized for Seedance)
                    │
                    ├─► Gemini 3.1 Pro watches it ─► analysis.json (hook, script, person, place, beats, sound)
                    └─► cut detection (ffmpeg)     ─► shots.json + one frame per shot
                                   │
                your changes ─► direction.json / plan.json / settings.json
                                   │
          Replace the person                       Make a new version
          ──────────────────                       ──────────────────
          cast a new face (GPT Image 2.5)          cast a new face
          mute the clip (new voice)                Gemini describes every measured shot
                                                   GPT Image 2.5 redraws each shot fresh:
                                                   same framing, new people, new place, no text
                                   │
                prompt ─► the voice read from the new face, who says which line
                                   │
                Seedance 2.5 ─► zuverse-<project>.mp4
                                   │
                only if you ask ─► edit: Gemini times every word, ffmpeg measures the quiet,
                                   Claude writes edit.json ─► zuverse-<project>-edit.mp4

Some details that make the difference:

• Cut for cut. The cuts come from the frames, not from the model's guess (left alone it folds a 12-shot ad into 5), and the remake is told to cut at exactly those times.
• A new voice, really. Seedance copies a reference video's sound, voice included, so the clip is sent muted and the new person's voice is read from their face.
• Two voices stay two voices. A phone call or someone behind the camera keeps their own lines and voice.
• No text by default. AI video misspells captions, so on-screen words are left out (turn them on with text=on).
• Nothing traceable in a new version. Shot frames are drawn new (same angle, new people, a new place), never edits of the original frames, and the presenter is cropped the way the ad crops them.
• Cuts that never clip a word. The edit cuts a pause only where the word timings and the measured quiet agree, so a soft voice on a phone is never cut for silence. Captions are drawn by ffmpeg (real text, spelled right), not by the video model.

Troubleshooting

• Facebook Ad Library link won't download. Facebook shows those pages only to a real browser. Install yt-dlp, or open the ad, right-click the video, "Copy video address", and fetch that; or save the page and use fetch --html saved.html --ad <id>.
• content_policy. The video model declined it (free). Seedance's filter often declines photoreal faces in reference images, generated ones included. Try once more; if it's declined again, set images=off (a new version then sends no faces and the video model casts the person from the written plan, shot by shot), or switch a Replace to mode=new.
• ffmpeg not found. Run npm install, or install ffmpeg from your package manager.

Use it responsibly

Zuverse is for turning ideas that work into your own ads. Don't use it to impersonate a real person, put someone's face or voice in an ad without their consent, copy logos or trademarks, or pass off someone else's ad as yours. Follow the terms of the platforms you post on.

License

MIT. Made by the team behind Zuops and Zurelay.