How to make an AI time travel video
AI time travel videos took over Instagram — one account hit 547K followers in 21 posts. The trick is one detail almost everyone skips. Here's the recipe.
Your feed is full of modern girls being dropped into the Titanic. Most look fake. One account did it 21 times, hit 547,000 followers in under a month — and the trick is one detail almost everyone skips.
@chloe.vs.history — a Gen Z character in modern streetwear, dropped into Pompeii, the Titanic, Victorian London — pulled 4.3M+ likes on a single video with fewer than 15 posts.
The format only works because of character consistency — same face, same tattoo, same modern clothes across every era. Get that wrong and it reads as AI slop.
The whole recipe is five moves: lock the character, pick an event everyone already knows, write a Cassandra script, shoot it like a phone selfie, layer period-accurate sound.
An AI time travel video is a 9:16 selfie-style clip where a modern character lands inside a famous historical disaster and reacts to it like they're filming a vlog. The breakout case is @chloe.vs.history — a tattooed American girl wandering through Pompeii, the Titanic, and Henry VIII's court, eating Roman fast food, warning passengers about icebergs. The first post hit 136K likes with zero prior audience. The Titanic post, three days later, hit 329K likes and 3.7K comments.
Then in late May, Sky News pulled on the thread. Chloe isn't a girl. She's an AI character built by Jonathan Laramy, a British millennial in Gloucestershire who also runs Majestic Studios — a YouTube history channel that did 14M views in 90 days using the same playbook. The audience, told the host doesn't exist, mostly shrugged.
Receipts: Think Like a Creator on the Chloe reveal (May 26, 2026); Alici.ai's reverse-engineer of the formula; Vuela.ai on the 547K-in-21-posts breakdown.
How to make an AI time travel video that works
Lock the character before you generate anything else. Build a 6–8 image reference sheet in Nano Banana 2 of one modern person: same face, same hair, same one or two identity markers (a visible tattoo, a band tee, a gold chain), in different poses and lighting. This is the asset that travels with you into every era. If the face drifts even slightly between posts, viewers register it as AI and stop following. This step is the entire moat.
Prompt sketch: "Character reference sheet, 6 portraits of the same young woman, identical face, short black bob, small star tattoo behind left ear, modern streetwear (white tee, leather jacket, beige sweater, slip dress, denim jacket, hoodie), bright clean studio lighting, hyper-realistic, no setting variation."
Pick a historical moment everyone already knows. Titanic. Pompeii. Black Death. Henry VIII. Victorian London. The reason these work is that the audience already knows the ending — and that pre-loaded knowledge is what makes the video emotionally complete in 60 seconds. If you have to explain the event, you lose. The Battle of Jarnac will never beat the Titanic, even if the AI-generated visuals are better. Pick from the school-curriculum / blockbuster-film shortlist.
Write a Cassandra script — the protagonist tries to warn them, and fails. Chloe's Titanic post is the pure version: "I tried to warn them… but they didn't listen." The audience already knows the ship sinks, so dramatic irony does the emotional work. The escape variant is the mirror — survive the era and judge it on the way out ("4/10, wouldn't recommend Tudor London"). Both close the loop. A neutral tour does not — that's the format every low-engagement clone falls into.
Worked example: a modern guy in a band tee strolls through 1880s London, hyped about the gas lamps, then realises he can't get his phone to charge, then notices the smog, then the cholera. Closing line: "Get me out of here."
Shoot it as a 9:16 handheld vlog, not a cinematic short. The three words that lock the aesthetic in every video clip prompt: "handheld selfie, hyper-realistic, cinematic atmosphere." Vertical, medium close-up, slight natural jitter, direct address to camera. The polished cinematic version of this same idea performs worse — the selfie aesthetic is what makes the audience pattern-match to authentic UGC instead of an AI showreel. Drive the actual frames with a recent video model — Kling 2.6 Pro, Seedance 2 Pro, or Veo 3.1 — passing your character reference at every shot so the face stays locked.
Layer period-accurate sound, or the whole thing collapses back to AI slop. Horse hooves on cobblestones. Gas lamp hiss. Distant coughing. Squelching mud. Latin dialogue in the opening seconds of the Pompeii cut. Chloe's videos get away with hyper-real AI visuals because the audio is right. Generate the ambience layer in ElevenLabs, or pull from a royalty-free archive. One sound design pass per video — non-negotiable.
The uncomfortable thing about the Chloe story isn't that an AI character built a million-follower account. It's that when the audience found out, they mostly shrugged and kept watching. Teachers were already using the videos in class. The format works because the format works — the host being a man in Gloucestershire didn't change the historical accuracy or the dramatic irony. Which means the cost of producing a creator is collapsing toward zero, and the creators who will still be standing in two years are the ones whose realness is unmistakable — or the ones who own a locked AI character of their own from day one.
The bottom line on AI time travel videos
The AI time travel format isn't a one-off Chloe trick — it's the first popular template that pays creators for solving the hardest problem in AI video: keeping one face consistent across every post. The five-move recipe above works on every era you can name, in every niche where the audience already knows the ending. The catch is step one. If your character doesn't survive between videos, the rest of the build doesn't matter. Lock yours in →
AI time travel video FAQs
What tools does Chloe VS History use?
The exact stack isn't public — the account doesn't publish its production files. The reverse-engineered recipe that matches the output uses Nano Banana 2 (or Nano Banana Pro) for the character reference sheet, Kling 2.6 Pro / Seedance 2 Pro / Veo 3.1 for video generation with character reference, and ElevenLabs for voice plus period-accurate ambient audio.
How long are the actual videos?
Around 60 seconds on Instagram Reels and TikTok, with the YouTube cuts running 2–4 minutes. The shorts open on the anachronistic contrast in the first two seconds, then move through 3–5 short scenes to a single payoff line direct to camera.
Does the character have to be human?
It has to be a recognisable identity. The viral accounts all use a single human character because the face is what the algorithm and the audience both learn to recognise. An anthropomorphised mascot can work, but you lose the anachronistic-contrast hook that does most of the heavy lifting.
Is it okay that the host doesn't exist?
The Chloe audience didn't punish the reveal. That's the data point. The accounts that get into trouble are the ones that pretend to be a real specific person and accept money on that basis. A clearly fictional AI presenter, doing accurate historical content, has so far been treated by audiences as fine.
How do I avoid the AI-slop label on this format?
Three things separate the clones from Chloe: rock-solid character consistency (same face, every scene), a script with a real emotional payoff (Cassandra or escape — not a neutral tour), and period-accurate sound design. Get all three right and the AI-generated visuals stop being the conversation.