SLS Camera (Ghost Vision)
SLS camera with AI pose estimation (TensorFlow Lite). Detects and tracks human figures with 17 keypoints in real time, like a professional SLS camera. How an SLS camera works →
Deep Learning
MODULES
Ghost Hunting Tools
Each module is an independent analysis tool, calibrated to maximize the detection of paranormal anomalies using your phone's real sensors.
AGSL Night Vision
Professional GPU-processed filter. Boosts ISO gain and optimizes the image in low-light conditions for night investigations.
GPU Processing
Precision EMF Radar
Measures electromagnetic fluctuations using your device's native magnetometer with a dedicated EMF radar interface in microteslas (µT).
Magnetometer
Digital Geophone
Detects micro-vibrations in surfaces through high-sensitivity accelerometers. Perfect for static structures and séance tables.
Accelerometer
Waterfall Spectrogram
Real-time audio frequency analysis. Visualize and detect sound anomalies with a waterfall display for EVP sessions.
Audio FFT
EVP Audio Engine
Captures and analyzes Electronic Voice Phenomena. A dedicated audio engine for pattern detection across the sound spectrum.
EVP Analysis
24/7 Surveillance Mode
Leave the phone fixed and let it monitor on its own. Continuously logs everything the sensors and camera capture through the night.
Static Monitoring
Gallery with Filters
Review your captures and apply negative, cold, warm, edges, histogram filters plus the ghost modes EVP, cold and spectral, with brightness, contrast and saturation control.
Image Analysis
Dead Hour Alert
Set smart alerts for the Witching Hour (03:00 AM), the time of peak reported paranormal activity. GhostTech starts the investigation automatically.
Smart Alert
How it works
Uses the camera together with a TensorFlow Lite pose-estimation model that runs inside your phone. It looks for 17 key points of the human body — shoulders, elbows, hips, knees — in every frame and, when enough of them line up, it draws a skeleton over the image. Same idea as research SLS cameras: it doesn't see ghosts, it sees shapes that resemble a human body.
How to use it
- Turn the camera on and hold the phone still for a few seconds: the model needs stable frames.
- Adjust the sensitivity. Low = few figures, more reliable. High = many figures and many false positives.
- Frame the whole empty space, with no people or reflections, and let it run.
- When a skeleton appears, capture: the photo is saved linked to the sensor readings from that exact moment.
Warnings and limits
- The skeleton is not proof of anything. The algorithm is trained to find bodies and sometimes forces one onto coat racks, chairs, plants, curtains or shadows.
- A figure that flickers for an instant, or has impossible proportions, is almost always a false positive.
- At high sensitivity you'll see figures constantly. That doesn't mean more activity: it means a less demanding filter.
Phone requirements
- Camera and camera permission (asked only when you switch it on, not when you open the app).
- The model runs on your device: the image never leaves the phone unless you request the AI analysis.
- On older phones the processing drops to fewer frames per second.
Read more on the blog
How it works
An AGSL shader processed by the GPU that raises the gain and re-stretches the image range to lift detail out of near-darkness. It is not thermal or infrared vision: it amplifies the little light already reaching the sensor.
How to use it
- Turn the lights off and wait a minute: both the sensor and your eyes need to adapt.
- Rest the phone on something fixed. With the gain up, any shake shows twice as much.
- Raise the filter gradually, until you can see the back of the room without washing out the detail.
- Combine it with the SLS camera to sweep dark spaces.
Warnings and limits
- More gain means more digital noise. That grain creates dots and blotches that look like orbs or figures, and aren't.
- It does not replace an infrared light source: in absolute darkness there is nothing to amplify.
- A heavily pushed image loses real detail. Don't base a verdict on it alone.
Phone requirements
- Camera and a GPU with AGSL support (the full shader needs Android 13 or newer).
- All filtering happens locally, on your device.
Read more on the blog
How it works
Reads your phone's native magnetometer — the same chip that drives the compass — and shows the magnetic field in microteslas (µT) on a radar interface. GhostTech flags any spike above 60 µT as an anomaly.
How to use it
- Do a calibration pass first: walk the room and note the baseline reading in every corner.
- Stay away from sockets, routers, fuse boxes, speakers, the fridge and your own charger.
- Look for sudden changes, not high steady values. A spike that comes and goes tells you more than a big constant number.
- Repeat the same walk several times: whatever repeats in the same spot almost always has a physical cause.
Warnings and limits
- Nearly every spike has a mundane source: wiring, speaker magnets, motors, metal pipes, the phone itself.
- Earth's magnetic field already sits around 25-65 µT depending on where you are. A 50 µT reading on its own means nothing.
- Magnetic-clasp cases and car mounts throw the measurement off completely.
Phone requirements
- A magnetometer. Not every budget Android has one; without it the radar cannot measure.
- No permissions and no internet connection needed.
- Pro module.
Read more on the blog
How it works
Turns the accelerometer into a tabletop seismograph: it measures surface micro-vibrations in m/s². First the gyroscope checks that the phone really is still (below 0.05 rad/s); if it isn't, the reading is flagged invalid.
How to use it
- Rest the phone on a firm surface: table, floor, shelf. Never in your hand.
- Step a couple of metres away and stay still. Your own footsteps are the number one source of false positives.
- Wait for the indicator to confirm the phone is static before giving any reading weight.
- Log spikes above 2 m/s² with their timestamp and cross-check them later against household noises.
Warnings and limits
- If the phone is not still, a 0 does not mean "no vibration": it means "invalid reading". Don't interpret it.
- Lorries, roadworks, lifts, the washing machine, the heating and wind on the windows move the floor perfectly normally.
- On a soft surface — bed, sofa, thick rug — the geophone loses almost all of its usefulness.
Phone requirements
- Accelerometer and gyroscope.
- A rigid, stable surface to rest it on.
- Pro module.
Read more on the blog
How it works
Analyses microphone audio at 44.1 kHz with an FFT and draws it as a waterfall: time falls downwards, frequency runs horizontally and colour marks the energy. It lets you see the shape of a sound instead of only hearing it. On capture it freezes the last 96 frames.
How to use it
- Record in silence first and study the background band: that's the normal noise of the place.
- Look for traces that stand out from that background, especially in the human voice band (300 Hz - 3 kHz).
- Note the timestamp of every odd trace so you can go back to the original audio.
- Always cross-check what you see against the geophone and EMF from the same moment.
Warnings and limits
- A pretty trace is not a voice. Most are pipes, the fridge, traffic, wind or the mic clipping.
- Your phone's microphone applies its own noise reduction: it can manufacture artefacts that look like signal.
- Seeing a pattern is not understanding it. The interpretation is yours and always debatable.
Phone requirements
- Microphone and recording permission (asked only when you start the analysis).
- Pro module.
Read more on the blog
How it works
Captures raw microphone audio at 44.1 kHz and analyses it live, looking for patterns in the spectrum. This is the module for EVP sessions: it records, saves and links every capture to the exact moment of the session, alongside the sensor readings.
How to use it
- Pick the quietest spot you can and state out loud the time, the place and who is present.
- Ask short questions and leave 10-15 seconds of clean silence between each one.
- Call out any noise that happens ("a car just went past"). That saves the recording when you review it.
- Listen back with headphones, without knowing in advance what you expect to hear.
Warnings and limits
- Auditory pareidolia is extremely powerful: if someone tells you which word "you can hear", you will hear it. Listen blind first.
- Background noise, radio interference, the case rubbing and your own breathing produce most alleged voices.
- Audio is not proof. Without a second independent recording of the same moment, there is nothing holding it up.
Phone requirements
- Microphone and audio recording permission.
- Headphones for review: the phone speaker swallows the detail.
Read more on the blog
How it works
Leave the phone fixed in place and it watches on its own. Four observers run in parallel — camera, EMF, geophone and audio — and each logs an event when it crosses its threshold: figure detected, EMF above 60 µT, vibration above 2 m/s² or an audio peak. Everything lands on a timeline with the exact time.
How to use it
- Pick which sensors you want active and name the session.
- Set the phone down plugged in, fixed and framing the area. The mode pauses if you leave the app for too long.
- Let it run all night and review the timeline in the morning.
- Cross-reference events: what matters is several sensors landing in the same minute, not one lone spike.
Warnings and limits
- Events are throttled to one every 5 seconds. If something repeats a lot, you'll see fewer entries than actually happened.
- A whole night generates a huge number of trivial events: temperature shifts, the heating, traffic.
- It drains the battery and heats the phone. Leave it charging and ventilated.
- Local data auto-deletes after 30 days.
Phone requirements
- Whichever sensors you enable, each with its own permission.
- A charger and a fixed surface.
- Pro module.
Read more on the blog
How it works
Review your captures and apply image-analysis filters: negative, cold, warm, edge detection, histogram and the EVP, cold and spectral ghost modes, with fine control over brightness, contrast and saturation. Every photo keeps the sensor readings from the moment it was taken.
How to use it
- Start with the histogram: it tells you whether the photo is blown out or underexposed before you draw conclusions.
- Use edges and negative to tell a real object apart from a compression artefact.
- Always compare against another photo of the same framing without the anomaly.
- From here you can request the AI analysis of a specific capture.
Warnings and limits
- A filter reveals nothing hidden: it rearranges what is already in the pixels. Pushing contrast manufactures figures.
- Compressed photos — JPEG, screenshots forwarded through messaging apps — create blocks and halos that look like anomalies.
- If you request the AI analysis, that photo and its readings do leave the device for the AI service.
Phone requirements
- Free space on the device for the captures.
- Cloud upload is optional and only happens if you ask for it.
Read more on the blog
How it works
Schedule an alarm for the so-called dead hour or Witching Hour (around 03:00) and GhostTech starts the investigation for you at that exact time. It is a logistics aid, not a claim that anything special happens at that hour.
How to use it
- Set the exact time and leave the phone plugged in and positioned before you go to sleep.
- Prepare the framing and the sensors at night, not half-asleep at 3 AM.
- Take advantage of the house and street being at their quietest: that part is a real advantage.
- Note how you feel when you wake up. Tiredness heavily changes what you think you perceive.
Warnings and limits
- There is no evidence that paranormal activity increases at 3 AM. What changes is the silence and your own state.
- Waking abruptly out of deep sleep causes disorientation and sometimes sleep paralysis. That's physiology, not a presence.
- If cutting your sleep affects you, don't do it. No investigation is worth weeks of sleeping badly.
Phone requirements
- Exact-alarm permission and enough battery.
- Leave the phone charging: the alarm can't die halfway through the night.
Read more on the blog
UNDER THE HOOD
Cutting-Edge Technology
POSE ESTIMATION · LIVE
01
TensorFlow Lite — On-Device
Deep Learning models run locally. No network latency, no external dependencies. The MoveNet model detects 17 skeletal keypoints at over 30 FPS.
02
AGSL — Android Graphics Shading Language
Night vision shaders run on the GPU. Real-time image processing with hardware-level low-light signal amplification.
03
Edge Computing — Privacy by Design
The EMF radar, Ghost Vision, night vision, geophone and EVP engine are processed 100% on your device. Only when you request AI analysis of a photo are that image and its associated sensor readings sent, on-demand, to Gemini (Google) to generate a verdict — never in the background, never without your request. More details in our privacy policy.
STATIC MONITORING
Surveillance Mode
Leave your phone fixed anywhere and turn it into a paranormal security camera. GhostTech watches the environment for you, hour after hour.
- ✓ Continuous monitoring without having to hold the device.
- ✓ Automatic logging of every event detected by the camera and sensors.
- ✓ Combines EMF, geophone, EVP audio and Ghost Vision in one unattended session.
- ✓ Perfect for night investigations and empty rooms.
REC · SURVEILLANCE MODE
IMAGE ANALYSIS
Gallery with Paranormal Filters
Review each capture and apply analysis filters in real time. Reveal details invisible to the naked eye and fine-tune the image.
ORIGINAL
Ghost Mode Filters
Adjustments
PRO
PRO Mode
Unlock advanced tools and continuous monitoring through a simple reward system. No hidden subscriptions.
- ✓ Unlimited continuous monitoring
- ✓ All modules unlocked
- ✓ Advanced sensitivity controls
- ✓ Session export
12
3
6
9
The Dead Hour
03:00 AM — Witching Hour
Set smart alerts for the time of peak reported paranormal activity. GhostTech starts the investigation automatically.
INTERFACE
Built for Investigators
A technical, immersive interface with adjustable sensitivity controls to minimize false positives.
INFORMATION
Ghost Vision SLS›
Night Vision›
Waterfall Spectrogram›
Digital Geophone›
EMF Radar›
EVP Audio Engine›
Dead Hour Alert›
Privacy & Safety›
CLOSE
Information Panel
EMF
33.4 µT
PAREIDOLIA SENSITIVITY: 0.12
Main View — SLS + Geophone
EMF RADAR
31.1 µT
SCANNING...
Live EMF Radar
Privacy Done Right
Sensor and camera detection happens locally on your device; cloud AI only activates when you ask for it.
EMF radar, camera and audio 100% on-device
AI only analyzes on your request
We never sell your data
No recording without your permission
Anonymous usage analytics
FAQ
All About Ghost Hunting
What is an SLS camera and how does it work on a phone?
SLS stands for Structured Light Sensor. The original SLS cameras used the Kinect depth sensor to draw a skeleton over human-looking shapes. Since phones don't have that infrared sensor, the SLS camera in GhostTech (Ghost Vision) reproduces the effect with AI pose estimation: a TensorFlow Lite model looks for human silhouettes in the image and marks up to 17 keypoints in real time, all processed on your own device. Full guide to SLS cameras →
Is the SLS camera free in GhostTech?
Yes. The Ghost Vision SLS camera is included free when you download GhostTech on Google Play, alongside the EMF radar, EVP recorder, night vision, geophone and surveillance mode. You don't need external hardware or a subscription: your Android camera is enough.
Why does the SLS camera mark figures where nobody is standing?
Because the algorithm is trained to find human bodies and sometimes forces a skeleton onto objects with a similar silhouette: coat racks, chairs, plants, shadows or reflections. These are false positives and they are normal in any SLS camera. A figure that flickers for an instant, or has impossible proportions, is almost always one of them.
How does a ghost detector app work on a phone?
GhostTech uses your Android's real sensors — magnetometer, accelerometer, microphone and camera — to measure electromagnetic fields (EMF), micro-vibrations, audio anomalies and movement. That data is processed with AI directly on the device to flag possible anomalies during a paranormal investigation.
What is surveillance mode and what is it for?
Surveillance mode lets you leave your phone in a fixed spot so it monitors the environment continuously. It automatically logs every event the sensors and camera capture, ideal for night investigations and empty rooms with no direct supervision.
What is an EVP?
EVP (Electronic Voice Phenomena) is the supposed recording of voices or sounds that are not heard at the moment of capture. GhostTech's EVP audio engine and waterfall spectrogram help you record and analyze the sound spectrum for patterns.
What filters does the gallery have?
The gallery includes negative, cold, warm, edges and histogram filters, plus the ghost modes EVP, cold and spectral. You can also manually adjust brightness, contrast and saturation of each capture to reveal details.
Is GhostTech free and does it need internet?
GhostTech is a free download on Google Play. All AI, camera and sensor processing happens locally on your device, so it works without sending data to the cloud or needing a constant internet connection.
Does GhostTech really detect ghosts?
GhostTech is a tool for entertainment and technical experimentation. It measures real signals from the device's sensors, but results should be interpreted critically: an electromagnetic or sound anomaly is not scientific proof of paranormal activity.
⚠ SCIENTIFIC NOTICE
GhostTech is a tool for entertainment and technical experimentation. The results obtained should be interpreted with critical, scientific judgment. Anomaly detection depends on the quality of your device's sensors and the conditions of the environment.
YOUR VOICE MATTERS
What would you like to hunt next?
Tell us what feature, sensor, or experiment you'd like to see in GhostTech. We read every suggestion.
