← All posts
Jun 26, 2026 · ShortRemix Team · 4 min read

The tool we built to recreate viral YouTube Shorts scripts

Why generic AI script generators fail at remaking viral shorts, and how ShortRemix's structural extractor produces remake scripts that hold the original's pattern.

I spent most of 2025 doing the work in the structural extraction post by hand. Scrub through a winning short. Note what's on screen at each second. Note what emotion the viewer is in. Map the beats. Pick the angles. Write the scripts. Write the editor brief.

It took about 90 minutes per video to do well. The output was three scripts, an editor handoff document, and an honest read on whether the pattern would transfer to my niche.

After roughly 40 of those, I started building a tool to do the slow parts. That tool became ShortRemix. This is what it does, why generic AI script generators don't, and how to try it.

What "recreating a viral short" actually requires

If you ask a generic AI script tool to "remake this viral YouTube Short," you get something that sounds like the original. It rephrases the words. It keeps the topic. It usually invents a slightly worse hook line.

That output is useless because it copied the surface. What you have to copy to make a remake actually land is the structure — the hook type, the beat map, the energy curve, the setup-to-payoff ratio, the emotional sequence the viewer goes through.

The generic tool doesn't carry that representation. It has the script as text. So all it can rotate is the words.

ShortRemix builds the structural representation first, then generates scripts from the structure — never from the original text. The output is scripts that share the original's mechanic but use entirely different content for your niche.

What you get when you drop a short in

You paste a YouTube Shorts URL or upload a video. About 60 seconds later you get:

  • A second-by-second structural map: what's on screen, what the speaker is saying in summary, what emotional state the viewer is in.
  • The hook pattern named explicitly — belief flip, list with missing number, prop reveal, disqualifier, cold open, or something less common if the video doesn't fit one of these.
  • A beat map of 5-second segments labeled by the information event that lands in each.
  • Three to five remake scripts in your niche, using the same structure but completely different content.
  • An editor brief in the format I described here, with the "what this is not" section pre-populated for your niche.
  • A shareability score per script so you can filter which to record.

That's the pack. You read it in five minutes. You pick the scripts that fit. You record.

Why generic AI tools fail at this specifically

Most "AI YouTube Shorts script generators" do one of two things.

They start from a topic prompt and write a 30-second script. The script is generic because the model has no structural reference — it's optimizing for "sounds like a YouTube Short" not "uses a hook pattern that's been validated on this specific video."

Or they start from a transcript and "summarize" or "rephrase" it. The output is a paraphrase, not a remake. Viewers can tell within three seconds.

Neither workflow produces scripts you'd actually publish. The pattern-extraction step is the entire missing piece, and it's the step that takes the most expertise — which is why most tools skip it.

What's actually different

Three things, concretely:

The extraction is structure-first. The script generator never sees the original text — it only sees the structural map. This forces fresh content with preserved mechanics.

The output is editor-ready, not "looks like a script." The brief format is the one that prevents 80% of revisions in real handoffs.

The angle generator runs the five-axis rotation I described in the ten-scripts post — audience swap, contrarian flip, time scale, format swap, specificity zoom — so the output is varied without being random.

Result: scripts you can actually publish, in roughly the time it takes to read a long email.

Who it's for

Creators publishing shorts who already have an eye for what works on their channel but don't have time to do the structural breakdown by hand for every winning video they want to learn from.

Agencies running multiple client channels, where the workflow has to be repeatable across niches.

Editors who get briefs from creators that are too vague — the pack output gives them the missing constraints.

It's not for people who want generic AI to write their videos for them. It's for people who already understand what makes shorts work and want to scale that understanding across more videos than they could analyze by hand.

How to try it

The product is in private beta. Pricing is published here — pick a plan to signal interest. Beta access opens in waves; plan-interest choice tells us who to invite first.

If you want to see what the output actually looks like before signing up, the example pack walks through a real short end to end. That's the entire workflow rendered for one video.

If you want to do the workflow by hand, the recreate-without-copying post is the methodology. The tool just makes it faster once you have the eye for it.