My haunted mirror possesses your reflection with open-weight AI. A player’s idea made it hunt you in the dark
This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine After trying my haunted mirror, a player sent me the most useful message I got all week: "Instead of looking at yourself, make the screen go black. Say it can't see you. Then jump out with a really loud noise." They were right. If you can watch it coming, you're ready for it. So I built lights-out…
This is a story about an AI-powered haunted mirror called Scare-TV that uses open-weight models to scare people in various ways. The developer built Scare-TV after a player suggested turning the screen black and adding a loud noise to enhance the scare effect. The resulting "lights-out mode" resulted in 44% more scares on average compared to the regular mode.
Scare-TV uses a laptop with a webcam and a TV screen to project different ghostly faces onto a person's reflection. These faces include black eyes, a mouth that opens wide, a pale double behind the person, and a grin that isn't their own. The scare can also include a shadowy girl peeking from behind the person's shoulder, melting, or a grimace that isn't their own.
For the phone version, Scare-TV uses a front camera and MediaPipe Face Mesh to detect faces. It then runs a second open model called SD-Turbo to generate realistic horror faces on the user's device. YOLO-pose then grades these faces to ensure they meet certain criteria before allowing them to scare the person.
Scare-TV also offers 20 classic jump scares for those who don't want to use their camera. The server records each person's scare count, the highest jump level achieved (0-10), and whether they ran out of frame. This data is stored in a PostgreSQL database on Render, where a Django admin interface allows the developer to filter data by various parameters. The developer's goal was to create a horror experience that doesn't require any copyrighted characters or pre-recorded samples.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
