AI Mystery Video Generator

Dark Mystery, the default here, is moody, foggy and suspenseful: a calm sea under low cloud, a runway at dusk, a forest road at night. It suits lost ships and planes and unexplained places, where the setting carries the unease. Its scenes look like photographs, so ask for places, weather and objects rather than people, and treat every scene as an illustration: a realistic picture of a real event is exactly what the AI rules below are about. A video in it typically costs about 25 credits.

Vintage Archive renders each scene as an old black-and-white photograph, which dates a case at a glance. It suits the 1850s to the 1960s, from Earhart's last flight in 1937 to Flight 19 in 1945. It's also the look most easily mistaken for evidence: an AI 'photo' of Earhart's plane is a new picture, not an archive one. Keep faces out, and say in the narration or the post text that the images are AI reconstructions. About 25 credits.

Documentary stock footage is the cheapest look, at about 12 credits a video: licensed clips of real seas, coastlines, forests and old buildings, with nothing to attribute. It shows the place today, not the event. When no clip fits a line, the pipeline falls back to an AI image, so a footage video can still contain AI scenes; check every scene before you post.

Pencil Sketch draws each scene in graphite on paper, the look of a case file: a logbook, a hand-drawn chart, a page of notes. It's plainly a drawing, which makes it the honest choice for reconstructions, and for a puzzle like the Voynich manuscript, where the object itself is the mystery. About 25 credits.

Classical Painting turns every line into an old-master oil painting, the look for mysteries older than photography: an ancient puzzle, or the Roanoke colony, the 117 men, women and children John White led to the New World in 1587, who had disappeared when he returned to Roanoke Island three years later. Nobody mistakes an oil painting for a photograph. Also about 25 credits.

Every look here except stock footage can also turn each image into a short AI video clip. That raises a typical video to about 464 credits, so keep it for a series' strongest cases rather than the daily post.

"Flight 19: on 5 December 1945, five US Navy Avenger torpedo bombers left NAS Fort Lauderdale on a training flight over the Bahamas; the 14 men aboard were never found. Use the Navy's record: the leader radioed that both his compasses were out; the Navy's final finding: cause unknown. Dark Mystery look, sea and storm, no faces, calm voice. Tell the Bermuda Triangle as a legend, not a cause."

"Amelia Earhart's last flight: on 2 July 1937 she and navigator Fred Noonan left Lae, New Guinea, for tiny Howland Island, where the Coast Guard ship Itasca waited with fuel. It heard her last at 8:43 a.m.; they never made two-way contact. Vintage Archive look: the plane, the sea, the ship, no faces. End on what's unknown, and the 2024 sonar 'find' that turned out to be a rock."

"The Voynich manuscript, MS 408 at Yale's Beinecke Library: an illustrated book in an unidentified script by an unknown author, bought by bookseller Wilfrid Voynich in 1912. Tell it as a case file: what the library says is known, theories as theories, and why attempts to decipher it have failed, if it's a code at all. Pencil Sketch look, curious narrator. End on the scans anyone can view online."

Each fits the 400-character limit of the box above, and each rests on a source you can name: the Navy's history command, the Smithsonian's National Air and Space Museum and Yale's Beinecke Library. Write the documented facts into the prompt yourself, say what to do with the theories, and tell it to end on what's still unknown.

A mystery video is honest when it keeps two lists apart: what the record documents, with its source (an official investigation, a museum, an archive), and what people have suggested since. Say which list each line comes from. A theory can be the most interesting part of the video; it only has to be called a theory, never told as fact.

Flight 19 shows how much a good record holds, and where it stops. The Naval History and Heritage Command's account, which sets out to stick to the official record, gives the route, the weather, the 14 men aboard and the radio calls, including the flight leader's 'Both my compasses are out'. The Navy's board of inquiry blamed him; after his mother challenged the finding, the Navy cleared him, with the conclusion 'The cause of the accident remains unknown.' The account calls ditching in a fast-moving storm the most likely explanation, and says the loss 'most certainly wasn't due to aliens or the Bermuda Triangle'. NOAA adds: 'There is no evidence that mysterious disappearances occur with any greater frequency in the Bermuda Triangle than in any other large, well-traveled area of the ocean.' Tell the Triangle as the legend it is.

The Voynich manuscript shows how to handle a claim. Yale's Beinecke Library, which holds it as MS 408, says it was written 'in an unidentified script by an unknown author' and that cryptographic approaches have failed to decipher the text, 'if it is indeed encoded'. It calls the old link to Roger Bacon 'pure speculation', its catalogue notes that 'several scholars have claimed decipherments', and the library says it can't respond to individual theories, given how many it receives. So 'the Voynich manuscript, finally decoded' is someone's claim, not news: say whose it is, and that the library holding the manuscript still calls the text undeciphered.

Never invent new evidence, and don't let the pictures invent it for you. A realistic AI scene of wreckage on the seabed, a 'recovered' logbook or a 'decoded' page looks like a discovery, whatever the narration says. Show what's documented, such as the sea, the route or the object as it survives, and keep any reveal in the words. When a find is announced, wait for the people who hold the record. The Earhart prompt above ends on the 2024 sonar images a deep-water exploration company hoped showed her plane; the National Air and Space Museum reports it turned out to be a rock formation.

In the script, most of this is attribution: 'the Navy's board concluded', 'Yale's library says', 'one theory holds', 'no one knows'. Date the case in its first line, so nobody takes a 1945 disappearance for this week's news, and put the source in the post text, where viewers who ask 'source?' will look.

YouTube requires creators to disclose AI-generated or altered content that seems realistic, including content that 'alters footage of a real event or place' or 'generates a realistic scene that didn't actually occur', and content that makes a real person appear to say or do something they didn't. You answer Yes under 'AI use' when you upload; for videos ShortFast posts for you, the setting is in YouTube Studio. Non-realistic content needs no disclosure, and neither does using AI to write the script or the captions. YouTube says: 'Disclosing AI content alone won't limit a video's audience or impact its eligibility to earn money.'

Dark Mystery and Vintage Archive look photographic, so five Avengers over a stormy sea, or an 'old photograph' of Earhart's Electra, is a realistic scene of a real event: answer Yes. A footage video that fell back to a realistic AI image of a real place counts too, since YouTube's examples include 'AI generated extra footage of a real place'. Pencil Sketch and Classical Painting are plainly drawings and paintings. YouTube may also apply a label itself when its systems detect AI content, and creators who consistently don't disclose can face penalties, including removal of content or suspension from the YouTube Partner Program.

YouTube's misinformation policies are narrower than their name. They cover 'misleading or deceptive content with serious risk of egregious harm', with examples such as content 'technically manipulated or doctored in a way that misleads users' and content 'falsely claiming that old footage from a past event is from a current event', and YouTube may make exceptions for educational or documentary context. A video about a 1945 disappearance is a long way from those cases; on YouTube, the disclosure above and the monetisation rules below are the ones a mystery channel meets every day.

TikTok's rules reach further into this niche. Its Community Guidelines define conspiracy theories as 'beliefs about unexplained events, or claims that involve rejecting generally accepted explanations for events'. 'Conspiracy theories that assign blame to powerful groups or institutions, such as "the government" or a "secret society,"' are ineligible for the For You feed when they're likely to undermine public confidence or distort a matter of public importance, and conspiracy theories or hoaxes that could cause significant harm aren't allowed at all. A hoax, in TikTok's words, is 'false, fabricated, or misrepresented claims presented as facts'. 'The Navy covered it up' blames an institution; 'the Navy concluded the cause was unknown' is the record. TikTok allows 'documentary or educational content reporting on or condemning misinformation', which is where a video that tells a legend as a legend belongs.

TikTok also has rules for real events and real people. It doesn't allow misinformation that poses a risk to public safety or incites panic, 'including falsely presenting past crisis events as recent', and even with a label it doesn't allow AI-generated content that misleads about a matter of public importance, such as 'a crisis event, like a natural disaster or conflict' or 'content made to look like it comes from a real news source'. So date the case early, and never dress an old disappearance up as a breaking-news alert. TikTok also doesn't allow 'using the likeness of private figures without consent', and it counts everyone under 18 and every adult who isn't a public figure as a private figure: the crew of a lost plane, the passengers of a ship. Keep their faces out of the scenes and let the narration name them.

TikTok requires a label on AI-generated content containing realistic images, audio or video, and its guidelines give 'AI-generated images that look like real photographs or footage' as the example, which is what Dark Mystery and Vintage Archive scenes are. Artistic styles, like anime, and generic text-to-speech narration don't need one. TikTok encourages labeling AI-generated content and says the setting won't affect distribution, so turning it on costs nothing.

Labels and monetisation are separate questions. YouTube's monetisation policies list 'content that uses deceptive or misleading imagery or narratives' among what can't earn, and allow 'content that utilizes creative tools to assist in delivering a unique, well-researched, or creative narrative'. A daily series that tells every case to the same template risks the 'Generic or Repetitive Content' policy, which 'includes content that looks like it's made with a template, or that may feel repetitive to viewers after watching several videos in a row from the same channel'. In a mystery channel, the research is what makes each video different.

It doesn't verify claims. ShortFast writes a fluent script, and fluent isn't the same as documented: a draft can repeat a legend as fact, merge two cases or add a detail that sounds right and isn't. Nothing in ShortFast checks a date, a name or a quote, whether it writes the script or narrates yours, so check each line against the record before you render.

AI images of real places and people are interpretations. ShortFast draws what the script describes, not what Howland Island or a 1945 airfield actually looked like, and any face it draws is invented, whatever name the script gives it. Describe settings and objects rather than people, and if you have the rights to a real photograph, upload it into the scene. Yale has put scans of the whole Voynich manuscript online; check the rights statement on its catalogue record before you upload a page.

Stock footage isn't always footage. The documentary look builds scenes from licensed stock clips, but when no clip fits a line the pipeline falls back to an AI image. Check every scene, and treat the clips as the place today, not the event.

The narration is fixed once rendered. You can swap scenes, edit captions, change music, title and post text, and re-render, but a wrong date, or a theory said as fact, means generating again. There are no on-screen source cards either: the only text ShortFast adds is the burned-in captions of what the narrator says, so say the source aloud and put it in the post text.

An automation has no approval step. In a mystery series the AI picks each new case and writes its script, and every episode posts automatically or arrives by email, so choose email delivery and check each case against its record before you post it. The next two episodes are queued: you can edit their captions, scenes and post text, re-time, pin or delete them, but not change their script. The series brief, up to 500 characters, is where your rules go: documented cases only, theories named as theories, no cover-up claims.

None of these channels use ShortFast; they show what mystery audiences come back for. LEMMiNO ('Long-form videos about space and mysteries and whatnot') gives one case a long video, often 20 to 30 minutes and sometimes well over an hour, under titles such as 'The Legend of the Bermuda Triangle', 'The Lost Colony of Roanoke' and 'The Vanishing of Flight 370', with flat illustrations and maps on its thumbnails. Barely Sociable ('analysis videos for the odd corners of the internet') runs cases as a labeled series, such as 'The SOS Sign Incident - An Unsolved Mystery' and 'The Mysterious Black Box Of Flight 980 - An Unsolved Mystery', with grainy black-and-white thumbnails and bold red-and-white titles. Nexpo ('Exploring the Internet's dark side') names cases by place, as in 'Japan's Strangest Unsolved Mystery' and 'Siberia's Strangest Unsolved Mystery', over red-and-black illustrated thumbnails. Smithsonian Channel isn't a mystery channel, but its documentary clips include two-minute takes on classic cases, such as 'The True Story of the Mary Celeste' and 'How Roanoke Vanished Into Thin Air'.

What they share is one case per video and a title that says how sure the video is: a legend, an unsolved mystery, the true story. That is the part to copy; the scenes, the voice and the captions are the part ShortFast handles.