How do you make true crime videos with AI?

To make a true crime video with AI, pick a well-documented case, research it from court records and reputable reporting, write the script in chapters, choose one look and keep the same characters in every scene, then add a calm narrator, captions, a thumbnail and a title, and post it as part of a series. ClipFlux does the making in one tool: type the case, pick a look such as Noir Stickman, Mannequin 3D or Realistic, and it writes the script (or uses yours), draws every scene with the same characters, narrates it and adds captions, as a Short or a full-length 16:9 YouTube video in chapters. The part AI cannot do for you is the checking: every name, date and verdict needs a human against a real source.

Why true crime works on YouTube

A real case comes with everything a story needs: a mystery, people with motives, a mistake, and an ending that actually happened. Viewers stay to find out how it ends, and a channel never runs out of cases. Crime, heist and prison stories told with drawings or animation, with no presenter on camera, already pull millions of views:

The views were read on 28 October 2026, rounded the way YouTube shows them. None of these channels films a host or uses real crime scene footage: the look carries the video. That is what makes the genre a good fit for AI, and why the choice of look matters more than in most niches. The full table and the stick figure corner of it are in true crime stickman videos.

Step 1: Pick a documented case

Start with cases that are settled and well covered: famous heists, cons and frauds, prison escapes, gangsters, old cold cases that were finally solved. They have court records, books and years of reporting to check against. Search YouTube for the case first; if it already has a dozen recent videos, tell it from a narrower angle (the one clue, the one mistake, the witness nobody believed). Ongoing cases are the hardest to get right, because the facts are still moving and the people are still living through it. Cases to start from, each opening the builder in a look that suits it:

• How Al Capone was finally convicted for tax evasion (Vintage Toon)
• The Great Brink’s Robbery of 1950 in Boston (Vintage Toon)
• How Charles Ponzi’s scheme worked and how it fell apart (Doodle 2D)
• How the Unabomber was finally caught (Mannequin 3D)
• The Brink’s-Mat gold robbery at Heathrow in 1983 (Realistic)
• Who was the Somerton Man? (Creepy Stickman)

Step 2: Research it from court records and reporting

Read the court records where they are public, the major newspapers that covered the case and the trial, and books by reporters who followed it. Note what was proven, what was alleged and what is still a theory, and say which is which in the script ("police believed", "he was never charged"). Put your sources in the video description.

Then check the script line by line, whatever wrote it. AI writes crime stories fluently and swaps names, dates and verdicts with total confidence. Our own Mona Lisa heist video shows it: a 3:30 true crime story made with ClipFlux from one paragraph, 40 scenes in Mannequin 3D, shown unedited with the lines it gets wrong pointed out underneath. None of them sounds wrong when narrated. The full checking routine is in AI documentary videos.

Step 3: Write the script in chapters

A long true crime video works as a case file in five to eight chapters: the day it happened, the people, the plan, the investigation, the mistake, the trial, what happened after. Open on the strangest true detail in the first twenty seconds, then go back to the start. End each chapter on an open question so the viewer stays for the next one. For a Short, tell one case, one clue and one twist. Keep sentences short and tell it in the third person.

In ClipFlux you can type the case and let it write the script, or paste your own. When it writes a long video about a real subject, it writes from that subject's Wikipedia article, so the facts start from a source rather than from memory; you still check them. More on structure is in storytelling videos with AI.

Step 4: Choose a look and a cast

The look decides how the case feels before the narrator says a word: black and white for a detective story, a 1930s cartoon for a gangster, photo-real for a documentary. Pick one per channel and keep it, so every video is recognisably yours. The trap with AI pictures is drift: a character who looks different in every scene, or a style that slides halfway through. In ClipFlux every scene is drawn in the one look you picked, with the same characters, and any scene that comes out wrong can be redrawn with a note.

You do not have to invent the cast. ClipFlux has sixteen ready-made characters, and for crime stories the Detective, the Inmate and the Police officer do most of the work, with the Man in a suit and the Woman for everyone else. A picked character stays the same in every scene, and if you switch the video to another look that has the same character, it comes along drawn in that look. Why characters drift and how to stop it is in keeping AI characters consistent.

Which look fits which case

A rough rule: the closer the look is to a photo, the more care it needs. Noir Stickman, Mannequin 3D and the stick figure looks never draw a real face, which is exactly right when the people in the story are real. Realistic looks like real footage, so keep its people generic and never let a scene pass as a photo of the real crime. For the stick figure versions see true crime stickman videos and, for the eerie side, scary stickman stories. The faceless mannequin look is covered in the Fern-style guide.

Step 5: Add a narrator and captions

A low, calm voice that sounds like it is reading the file, never excited about the crime. Keep the same voice on every video so the channel sounds like one person. ClipFlux narrates the script with an AI voice and adds captions, which also help viewers keep names, places and dates straight.

Step 6: Make the thumbnail and title

The title names the case or asks the question ("How was the Unabomber caught?"); the thumbnail hints at the twist: one character, one object that matters, two or three words ("ONE MISTAKE", "NEVER FOUND"). Never put a real victim's photo on it. ClipFlux draws the thumbnail from the video's own characters and look, so it matches the video (see the thumbnail maker guide).

Step 7: Post it as a series

One video is a test; a channel is a series. Number the cases ("Case 001"), plan ten before you post the first, and mix the kinds: a heist, a con, an escape, a cold case, then repeat. Post on a steady schedule. A ClipFlux Channel can work through your list and post on the days and at the hour you choose (see YouTube automation with ClipFlux).

Real cases, real people

True crime is about people who were really hurt, and their families can find your video. Stick to what court records and reputable reporting say, call a suspect a suspect until a court says otherwise, and leave out private people who are not part of the public record. Show the moment before or after, an empty chair, an object left behind; never the violence itself.

Do not draw real suspects or victims as look-alikes, in any look. Use generic characters (the Detective, the Inmate, a Woman, a Man in a suit) and let the narration say who they are. A look-alike can be wrong, can look like someone else, and in a photo-real look can be taken for a real picture. YouTube also asks creators to label realistic altered or synthetic content at upload, which matters most for a photo-real retelling of real events.

YouTube's rules point the same way. Its violent or graphic content policy, as read on 28 October 2026, removes violence or gore meant to shock or disgust, and allows educational or documentary videos only when that context is in the pictures or audio of the video itself, not just the title or description. Its harassment policy bans mocking the death or serious injury of an identifiable person, denying or minimizing someone's role as a victim of a well-documented major violent event, and sharing private personal information.

How people make these by hand

The free tutorials teach a four-tool stack: ChatGPT for the script and the picture prompts, an image model for the scenes, ElevenLabs for the voice, and CapCut to cut it together. It works and costs little, but a ten-minute video needs a hundred or more pictures, and keeping the same detective, the same look and the same lighting across all of them is where most of the hours go. The whole manual workflow is in How to make AI videos with ChatGPT by hand.

Whichever way you make them, do not copy other channels' scripts. Since July 16, 2026 YouTube's monetization rules name videos that look made from a template or feel interchangeable as inauthentic content. Bring your own cases, research and angle. What YouTube pays for is in Can AI videos be monetized on YouTube?, long videos are covered in long-form faceless videos with AI, and the tool itself is the AI documentary video generator.

Questions