AI True Crime Video Generator

Dark Mystery, the default here, is moody, foggy and suspenseful, and ShortFast lists it for true crime, mysteries and horror. Its AI images can look like photographs, so use it for places, objects and atmosphere: a museum at night, an empty frame on a wall, a wet street under one lamp. Keep the faces of real people out of it, because a realistic AI face beside a real name looks like a photograph of that person. A video in it typically costs about 25 credits.

Pencil Sketch draws each scene in graphite on paper, close to a courtroom sketch, and it's the look for the people in a case: a gang around a table, a clerk at a counter, a jury. Nobody takes a pencil drawing for a photograph. Let the narration say who someone was, rather than asking the drawing for a portrait of a real person. About 25 credits.

Vintage Archive renders every scene as an old black-and-white photograph, which dates an old case at a glance: a Boston street in 1950, a 1949 truck, a 1940s shop window. It's also the look most easily mistaken for a document. Never ask it for a mugshot, a crime-scene photo or a 'police photo' of a real case, and say in the narration or the post text that the images are AI reconstructions. About 25 credits.

Comic Book ('Heists · Legends · Action stories') draws inked panels with halftone shading, which suits a heist or a con told as a plan: the casing, the keys, the night, the getaway. Stick Figures draws hand-drawn doodle characters, the look for a 'how the scheme worked' explainer, such as a code hidden in letters or the steps of a swindle. Both are openly drawings, and each costs about 25 credits.

Documentary stock footage is the cheapest look, at about 12 credits a video: licensed clips with nothing to attribute, for places and context only, such as a city street at night, a courthouse, a harbour. A stock clip is never the real crime scene, and the narration must not imply it is; when no clip fits a line, the pipeline falls back to an AI image. Every other look here can also turn each image into a short AI video clip, which raises a typical video to about 464 credits, and a moving Dark Mystery or Vintage Archive scene looks even more like footage.

"The 1990 Gardner Museum theft in Boston, from the FBI's account: on 18 March 1990 two men dressed as police officers rang the night bell, tied up the guards in the basement and spent 81 minutes taking 13 works of art. Dark Mystery look: the museum at night, the empty gilded frames, no faces. Name no suspects. End on the reward of up to $10 million the museum still offers."

"Boston's 1950 Brink's robbery as a heist in panels, from the FBI's account: in the months before, the gang slipped in after hours, took the lock cylinders from five doors one at a time, had keys made and put them back. On 17 January 1950 masked men took over $1.2 million in cash. Six years later one of them talked. Comic Book look, roles not names; end on the 1956 guilty verdicts."

"How a New York doll shop owner hid warship news in letters about dolls, from the FBI's case file: in 1942 she wrote to Buenos Aires in her customers' names, and a 'doll hospital' meant a shipyard. The letters came back 'Address Unknown' to the real women, who gave them to the FBI. Stick Figures look. Say she pleaded guilty to a censorship charge; the espionage charge was dismissed."

Each fits the 400-character limit of the box above, and each comes from an FBI history page you can name in the post text. Write the checked facts into the prompt yourself, say which words to use for who was convicted, and say what not to show.

Work from the record: an official account such as the FBI's history pages, court documents, or reputable reporting that names its sources. An AI draft can invent a name, a date or a quote, or attach the right detail to the wrong case, and in true crime each of those errors is about a real person. Check every detail against the source, then paste the corrected script: in script mode ShortFast narrates it word for word. Say the source aloud, and put it in the post text.

Use the record's own words for legal status: charged, accused, convicted, acquitted, pleaded guilty. The FBI's file on the doll letters shows why. The FBI describes the doll shop owner's scheme as espionage, but under a plea agreement the espionage indictment was dismissed and she pleaded guilty to a censorship violation, so a script that says she was convicted of espionage gets the record wrong. Never call someone guilty who wasn't convicted; saying a real person committed a crime the record doesn't support can also expose you to a defamation claim.

Real victims stay out of the AI: no AI image of a victim, no AI voice speaking as one, no graphic detail, and no opening on the worst moment. Tell the story knowing their family may watch it. Heists and cons are the easiest place to start, because the story is the plan: in the Brink's robbery, a gang that had keys made for five locked doors; in the Gardner theft, two men dressed as police officers ringing a night bell.

Prefer historical and closed cases. While a case is in court its facts are still being tested, and everyone accused in it is someone who hasn't been convicted, so that's where a mistake costs most. Describe people by their role, such as a guard, the gang, a passenger or a doll shop owner, and name no victim and no living person. For an open case like the Gardner theft, say only what the FBI says, and name no suspects.

YouTube requires creators to disclose AI-generated or altered content that seems realistic, and the first item on its list is content that 'makes a real person appear to say or do something they didn't do'. Two of its examples read like a true crime warning: 'depicting a public figure stealing something they did not steal, or admitting to stealing something when they did not make that admission', and 'making it look like a real person has been arrested or imprisoned'. You answer Yes under 'AI use' when you upload; for videos ShortFast posts for you, the setting is in YouTube Studio. Dark Mystery and Vintage Archive scenes can look like photographs, so when they show a real place or event, answer Yes; Pencil Sketch, Comic Book and Stick Figures are plainly drawings. Using AI for the script or the captions needs no disclosure, and YouTube says disclosing alone won't limit a video's audience or its eligibility to earn money.

A label doesn't make a scene acceptable: YouTube applies its Community Guidelines 'to all content on YouTube, including content that is AI generated or altered using AI'. Its harassment policy doesn't allow 'content that realistically simulates deceased or abused individuals describing their death or violence experienced', or 'content reveling in or mocking the death or serious injury of an identifiable individual'. So never script a real victim telling their own story, and never give one an AI face or voice.

Its violent or graphic content policy doesn't allow 'violent or gory content intended to shock or disgust viewers', and its examples include 'footage of crimes such as violent robberies that provide no education or explanation to viewers'. YouTube may make exceptions for educational or documentary context, but 'this context must appear in the images or audio of the video itself. Providing it in the title or description is insufficient.' In a ShortFast video, that context is the narration. For ads, its advertiser-friendly guidelines say 'content where the focal point is on blood, violence, or injury, when presented without other context, is not suitable for advertising'. An 'implied moment of death or severe bodily harm in an educational, documentary, journalistic reporting or music video context' can earn ad revenue, while 'a documentary on a recent homicide featuring descriptive language of the circumstances of death' earns limited ads.

TikTok requires creators to label AI-generated or significantly edited content 'that shows realistic-looking scenes or people', such as 'AI-generated images that look like real photographs or footage'; artistic styles, like anime, and generic text-to-speech narration don't need one. Even with a label, it doesn't allow 'using the likeness of private figures without consent', and it counts all people under 18 and every adult who isn't a public figure as private figures: in true crime, nearly every victim, relative and witness. It also doesn't allow 'sexualized, fetishized, or victimizing depictions', or AI content that misleads about a matter of public importance, including 'a public figure being degraded, harassed, or linked to criminal behavior'. TikTok encourages labeling AI-generated content and says the setting won't affect distribution, so turn the label on.

TikTok's graphic content rules don't allow real-world 'graphic incidents and violence, such as torture, the moment of someone's death, or the shooting of an individual', or dead bodies, and they keep 'the build-up to a serious injury or accident, even if the injury itself isn't shown' out of the For You feed. Its 'Violent and Criminal Behavior' rules allow 'documentary or educational content that raises awareness about violent and criminal behavior', but not 'promoting theft or the destruction of property' or 'sharing instructions on how to commit crimes that may harm people, animals, or property': tell how a heist happened, never how to repeat one. Its harassment rules don't allow 'degrading or revictimizing people who have experienced a tragedy, including claiming they deserved it or dismissing their experience'.

A daily true crime series has three more YouTube rules to stay clear of. Its monetisation policies list 'content that repeatedly uses disturbing themes (such as violence or loss) without building a cohesive narrative' among what can't earn, and its 'Generic or Repetitive Content' policy covers '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'. Its spam policy's 'Automated or synthetic mass-production' rule covers 'using automated tools or AI to churn out high volumes of similar content with minimal changes', while 'posting a few variations of a video is ok'. In true crime, the research is what makes each video different.

It doesn't check facts. ShortFast writes a fluent script, and an AI draft can invent a name, a date, an amount or a quote, or attach the right detail to the wrong case. Nothing in ShortFast checks them, whether it writes the script or narrates yours, so check every name, date and quote against the source before you render.

Dark Mystery scenes can look like photos. Never let one pass as a real photo of the case: a foggy street outside a real museum is an illustration, not the night of the theft. Keep faces out, say in the narration or the post text that the scenes are AI, and answer Yes under 'AI use' when they show a real place or event.

Stock footage is never the real place. The documentary look builds scenes from licensed stock clips that match the words, not the address, and when no clip fits a line the pipeline falls back to an AI image. Check every scene, and never tell viewers they're looking at the scene of the crime.

Small details can slip. ShortFast plans recurring characters, a shared setting and a palette before any image is drawn, but a uniform, a car or the number of people in a gang can still come out wrong in one scene. Describe them once near the start of the script, check every scene, and swap any that's wrong for another image or one you upload.

The narration is fixed once rendered. You can swap scenes, edit captions, change music, title and post text, and re-render, but correcting a wrong date, or 'convicted' where the record says 'charged', means generating again.

An automation has no approval step. In a true crime 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. Put your rules in the series brief, up to 500 characters: closed cases only, roles instead of names, no graphic detail, legal status in the record's words.

None of these channels use ShortFast; they show how much research one case can hold. All three make long videos. JCS – Criminal Psychology ('Forensic Psychology / True Crime / Social Science') puts stills from interrogation rooms on its thumbnails, under titles such as 'How To Interrogate a Narcissist' and 'What pretending to be crazy looks like'. fern calls its videos 'armchair documentaries' and takes on scams and investigations in titles such as 'How China's Biggest Scammer Got Caught' and 'Hansa: The Infiltration of the Dark Web'. Barely Sociable's investigations include crime online, in videos such as 'The Dark Side Of The Silk Road' and 'Inside The Illicit World Of iPhone Trafficking', under designed graphic thumbnails.

JCS's thumbnails come from interrogation rooms, and that is the one kind of image an AI scene should never imitate: a realistic AI 'interrogation room' or 'bodycam' frame of a real case is a scene that didn't happen. What all three share is one case or one subject per video, and a clear angle: psychology, how a scam worked, the internet. Copy that; ShortFast handles the scenes, the voice and the captions.