After Midnight: Turning Real Events into AI-Narrated Nightmares

Posted on Oct 31 • Edited on Oct 31 • AI-assisted

Before a door creaks or a shadow moves, those words have already done half the work. They make an ordinary room feel a little less ordinary.

That feeling inspired After Midnight, a horror-storytelling website I’m building around real-world inspiration, open-source AI, and voice narration.

The intended experience is simple: open the website, press "Tell Me Something Terrifying," and listen to a randomly selected nightmare.

Behind that experience, the website would search the web for unsettling historical incidents, reported mysteries, and unusual real-world events. An open-source language model would then use the retrieved material as inspiration for an original fictional horror story, which a narration component would read aloud.

"Based on a true story. The following is a fictional retelling inspired by reported events."

The goal is not to read a search result aloud. It is to turn a small, unsettling detail into a narrative that builds tension and leaves something unexplained.

Web source discovery ↓ Source review and factual notes ↓ AI-generated fictional horror story ↓ Voice narration ↓ Browser listening experience with source links Enter fullscreen mode Exit fullscreen mode 1. Find the spark A historical incident or an unusual report provides the starting point: an isolated location, an abandoned building, or something its witnesses could not explain.

Finding a claim online is not the same as verifying it. The source-review stage needs to distinguish documented events from legends and unsupported accounts.

The storytelling stage would build new characters, dialogue, atmosphere, and plot around that starting point.

I want the horror to come from ordinary details slowly becoming wrong—not just a stream of monsters and jump scares.

The narration stage would turn the generated text into an audio experience. Pacing and pauses are part of the intended effect: the silence before the next sentence can be as important as the sentence itself.

The following is an original fictional illustration, not a real incident or a demonstrated output from the website.

The hotel had been abandoned for thirty years.

Every night at 2:13, someone called reception from Room 307.

The demolition crew thought it was a wiring fault.

Until they disconnected the telephone lines.

And the next call came from the foreman’s pocket.

That is the kind of short, escalating unease I want After Midnight to create.

Creative control is the main motivation. I want room to experiment with narrative style, inspect the generation process, and choose models whose licenses allow the intended use.

Local inference is a possibility to explore, not a completed feature or a performance promise. Fresh web discovery would still require internet access, and any external search or narration service would have its own data-handling implications.

My interpretation is a machine that finds traces of the real world and gives fictional fears a voice.

The ghost is not something the website claims to detect. It is the unsettling story that emerges between the source material, the generated narrative, and the listener’s imagination.

The model and service choices still need to be confirmed in this write-up. Gemma is an open-weight model; listing other services here does not imply that those services are open source.

After Midnight is a work in progress. The next milestone is a demonstrable end-to-end experience: a sourced inspiration, a clearly fictional story, and playable narration.

Reality provides the spark. Fiction takes over.