Upload the reference ad to an AI agent and have it break the ad into named beats — each with a timestamp, location, narrative function, and emotional logic — then detect every cut, extract one frame per scene, and lock that shot breakdown before generating anything. In one documented run, the invideo agent detected 9 cuts automatically and preserved 7 shot beats across every recreated version.
Step 1 — Upload the ad and ask for a beat map. Give the invideo agent the reference video and instruct it to break the ad into named beats with timestamps, physical locations, the function each beat serves, and the emotional logic behind it — before any generation begins. invideo is an agentic video creation tool with all the current models available, and it reads uploaded videos directly: the agent watches the ad and comes back with a plan telling you what stays the same and what needs to change. Beat-level psychology matters as much as structure — for example, hook beats work emotion-first, product-second, so label what each beat does to the viewer, not just what appears on screen.
Step 2 — Detect every cut and pull one frame per scene. Instruct the agent to find the exact timestamp of every cut and extract one representative frame per scene — those frames become your recreation targets. In one documented recreation, the invideo agent detected 9 cuts in the reference ad automatically. The agent can also segment autonomously: in another run it watched a video, recognized an outfit change mid-ad, and split the footage into two clips without manual editing.
Step 3 — Transcribe the script and decode its structure. Have the agent transcribe the voiceover and identify the structural elements worth keeping — hook, body, CTA — as the template for your new script. It picks up linguistic devices too: one deconstruction surfaced that the reference was built on wordplay, riffing on a single sound throughout. If the reference is in another language, the agent can translate it and break down the music structure alongside the script.
Step 4 — Extract camera language and pacing into production rules. Ask the agent to pull the reference's visual grammar into structured headings — Visual Standard, Pacing & Camera, cut rate, camera style, on-screen text, energy, and ending pattern — and save them in the project context so every generation is governed by the original's style, not a generic default.
Step 5 — Review and lock the shot breakdown. The agent returns a structured breakdown table — shot number, duration, shot description, super text — derived from the reference. Review it yourself and lock it before any image or video generation; this is the gate that stops the recreation from drifting into generic structure mid-production. Ask for an estimated generation breakdown at the same step so you know roughly what you'll spend before committing credits.
Step 6 — Decide what stays constant and what changes. The edit structure, music bed, and pacing of a winning ad stay fixed; cast, location, voiceover, and on-screen text are the variables. Brief it plainly: keep everything identical, change only the character and the product — same camera, same pacing, same energy.
Step 7 — Recreate against the locked beats. Iterate still frames per beat first and spend video credits only on locked frames, then animate each locked frame — the invideo agent routes each shot to the right model, so Seedance 2.0 or Kling both run inside the same project without switching platforms. One caution: if the reference has burned-in captions, exclude the original video from generation prompts, or the model will reproduce those captions in your new footage — direct it from the extracted frames and locked breakdown instead. Stitch the finished clips in the same edit order as the original, and the recreation is done. Documented recreations of winning ads ran $30–$145 per variant depending on how much changed — about $70 per ad and ~2.5 hours each when batched across markets, with 8–10 variations of one winning ad achievable in a day once the deconstruction is locked.
Watch some of these to see what works for you:
Go through the reference ad, find where every cut happens, grab a frame from each scene. Redo all of them with the Nigerian woman, don't change anything else.
— invideo's creative team, on the exact prompt used for cut-level ad recreation