User Guide

Learn how to use our forensic analysis tools to detect altered, manipulated, or AI-generated media.

Video Tutorial

Watch our step-by-step guide to using clAIm

Video tutorial coming soon

To add your video, replace the placeholder below with an iframe embed

Science-Only Mode

Technical forensic analysis without bias

What is Science-Only Mode?

When enabled, Science-Only Mode analyzes media using purely technical forensic methods. The AI will not consider the source, credibility, or "plausibility" of the content. Instead, it focuses exclusively on detecting pixel-level artifacts, compression anomalies, and other technical signs of manipulation.

What it analyzes

Compression artifacts and double-compression
Clone/copy-paste detection
Splicing and compositing seams
Noise pattern inconsistencies
Editing software signatures
Resampling and scaling artifacts

What it ignores

Source credibility (e.g., "NASA photo")
Whether content "looks real"
Metadata claims about origin
Factual plausibility of the scene
Recognition of people/places

When to use: Use Science-Only Mode when you want to know if an image has been technically altered, regardless of its claimed source. This is particularly useful for analyzing media where the source claims high credibility but you want to verify the technical integrity.

Forensic Image Filters

Interactive tools to visually detect manipulation

Original

Detects: Baseline reference

View the unfiltered image as a baseline reference. Always start here to understand what you're analyzing before applying forensic filters.

How to use:

Use this as your starting point. Examine the image carefully for any obvious signs of manipulation before applying filters.

What to look for:

Obvious visual inconsistencies
Unnatural edges or boundaries
Color or lighting mismatches

Exposure/Levels

Detects: Compositing, splicing, tone mismatches

Applies extreme contrast adjustment to reveal hidden inconsistencies in lighting and tone between combined images. Manipulated areas often have slightly different tonal ranges.

How to use:

Drag the intensity slider to increase contrast. Look for areas that appear at different brightness levels or have inconsistent tonal response.

What to look for:

Regions that brighten/darken at different rates
Visible seams between composited areas
Inconsistent shadow density
Areas with clipped highlights or crushed blacks

High Pass

Detects: Selective sharpening, clone stamps, splicing

Highlights edges and sharpness differences across the image. When parts of an image are sharpened differently (common after editing), this filter reveals the inconsistency.

How to use:

Apply the filter and examine edges throughout the image. Areas that were edited often have sharper or blurrier edges than surrounding authentic content.

What to look for:

Edges with different sharpness levels
Halos around inserted objects
Repeated edge patterns (clone stamp)
Unnaturally sharp areas in otherwise soft images

Gaussian Blur

Detects: Healing brush, content-aware fill, smoothing

Averages pixel values to reveal over-manipulated areas. Regions that have been heavily processed (healing, content-aware fill) often respond differently to blur than authentic areas.

How to use:

Apply blur and look for areas that appear more 'mushy' or retain more detail than surrounding regions.

What to look for:

Areas that blur differently than surroundings
Texture inconsistencies becoming visible
Artificial smoothness in skin or surfaces
Patchy areas from content-aware fill

Edge Detection (Sobel)

Detects: Cutouts, masking artifacts, feathering issues

Uses Sobel edge detection to find all edges in the image. Jagged, unnatural edges often appear where subjects have been cut and pasted. Color-coded by direction.

How to use:

Examine the colored edge map. Look for edges that don't match their surroundings or have unnatural characteristics.

What to look for:

Jagged or pixelated edges around subjects
Double edges from poor masking
Missing edges where they should exist
Unnaturally smooth curves (over-feathered)
Cyan = horizontal, Magenta = vertical, Yellow = diagonal

Noise Analysis

Detects: Noise reduction, AI generation, splicing

Amplifies and visualizes the noise pattern across the image. Different cameras produce different noise signatures, and edited regions often have inconsistent noise.

How to use:

Look at the color-coded noise map. Red indicates high noise, yellow medium, and blue-green low noise areas.

What to look for:

Regions with different noise levels (different sources)
Unnaturally clean areas (noise reduction applied)
AI-generated content often has uniform, artificial noise
Spliced areas may have mismatched noise grain

Pixelate/Mosaic

Detects: Clone stamp, stretching, scaling artifacts

Breaks down the image into large blocks to expose stretched, scaled, or cloned pixel patterns that aren't visible at normal resolution.

How to use:

Increase the block size and look for repeating patterns or areas where blocks don't match their neighbors.

What to look for:

Repeated color patterns (clone stamp)
Stretched or distorted block shapes
Areas with different block characteristics
Interpolation artifacts from scaling

Lighting Analysis

Detects: Compositing, inconsistent shadows, lighting direction

Creates a gradient map showing light direction across the image. Composited elements often have lighting that doesn't match the rest of the scene.

How to use:

The color indicates light direction in each region. Authentic photos should show consistent lighting direction throughout.

What to look for:

Areas with different color (different light direction)
Subjects lit from opposite directions
Shadows pointing in wrong directions
Inconsistent highlight positions

Error Level Analysis (ELA)

Compression artifact visualization

How ELA Works

ELA re-saves the image at a known quality level and compares it to the original. Areas that have been edited will show different "error levels" than the rest of the image because they have a different compression history.

What to Look For:

Bright uniform areas: May indicate recently edited or AI-generated content (single compression pass)
Dark uniform areas: May indicate heavily compressed or multiple times saved regions
Inconsistent brightness: Different areas saved at different quality levels suggest editing
Edges with different levels: Spliced content often shows at boundaries

Note: ELA works best on JPEG images. PNG and other lossless formats may not show meaningful results. Also, images that have been re-saved multiple times may have uniform ELA patterns even if edited.

JPEG Forensics Panel

Deep file structure analysis

The JPEG Forensics panel provides detailed technical analysis of the image file structure. This is similar to tools like JPEGsnoop and can reveal editing software signatures, compression history, and metadata anomalies.

Overview

Shows the overall integrity score, estimated quality level, and whether double compression was detected.

EXIF Metadata

Camera make/model, software used, dates, GPS location, and other embedded metadata. Missing or inconsistent metadata is a red flag.

Quantization Tables (DQT)

The 8x8 tables used for compression. Different software uses different tables - this can identify Photoshop, cameras, etc.

Huffman Tables (DHT)

Entropy coding tables. Custom tables may indicate specific software or processing.

Software Detection

Identifies editing software like Photoshop, Lightroom, GIMP, or AI tools based on file signatures.

File Structure

Lists all JPEG segments (markers) in order. Unusual segment ordering or extra data can indicate tampering.

Audio Spectrogram

Frequency analysis for audio files

Reading the Spectrogram

The spectrogram shows frequency content (vertical axis) over time (horizontal axis). Brightness indicates intensity at each frequency.

AI-generated audio often shows:

Unnaturally uniform spectral patterns
Sharp cutoffs at specific frequencies
Missing natural frequency variations
Repeating patterns that indicate synthesis

Edited/spliced audio often shows:

Abrupt changes in the spectral pattern
Mismatched background noise levels
Discontinuities in harmonic content

Video Analysis

Frame-by-frame forensic examination

Video analysis extracts key frames and analyzes each one for signs of manipulation. The system looks for temporal consistency issues that indicate deepfakes or editing.

What the Analysis Detects:

Face swap artifacts: Boundary issues, texture mismatches, unnatural blending
Temporal inconsistencies: Flickering, jitter, or unnatural motion
Compression anomalies: Different compression levels between frames
Splicing evidence: Abrupt changes in lighting, color, or quality

Unmask: Beauty Filter Detection

Detect and visualize beauty filter applications

The Unmask feature analyzes photos for signs of beauty filter application and can generate an AI-estimated reconstruction of what the person might look like without those filters. This is useful for identifying heavily filtered profile photos or detecting cosmetic digital alterations.

What Filters Are Detected:

Skin smoothing: Artificial pore/texture removal
Face reshaping: Jaw slimming, eye enlargement, nose narrowing
Eye enhancements: Brightening, color changes, enlargement
Color adjustments: Skin tone evening, blemish removal
Virtual makeup: Digital blush, contouring, lip color

How It Works:

Step 1: Upload a photo (portraits work best)
Step 2: Click "Detect Beauty Filters" to analyze
Step 3: Review detected filters and affected areas
Step 4: Optionally generate an AI reconstruction

Important: Reconstructions are AI-generated estimations and may not accurately represent the actual person. They should be viewed as speculative illustrations showing what natural features might look like, not as definitive representations.

Pro Tips for Effective Analysis

Use Multiple Filters

No single filter catches everything. Use several filters together to build confidence in your findings. If multiple filters show anomalies in the same area, manipulation is more likely.

Compare Similar Images

If possible, analyze authentic images from the same source or camera to understand their baseline characteristics before analyzing suspected fakes.

Check the Metadata First

EXIF data can quickly reveal if an image was processed with Photoshop or other editing software. Missing metadata is also suspicious.

Consider the Context

Technical analysis is one piece of the puzzle. Consider the source, the claims being made, and whether the content makes sense in context.