Ape.I uses a 3-part multi-layered analytical approach:

01. Inspect Provenance

The system scans image content credentials for C2PA and JUMBF data. If the pixel metadata contains a verified technical signature this is the highest standard for authenticating image origination. 

Additionally, the Ape.I C2PA Inspector reveals the specific image source tool and the verified signer.

Hardware authentication will indicate a device such as a camera and the hardware manufacturer. e,g. Sony hardware capture, as well as the verified signer.

In the case of an AI generated image this will show details related to the specific source tool and parent organization. e.g. DALL-E Open AI (Chat GPT), Gemini Google LLC, and Firefly Adobe. With Ape.I this C2PA inspection is processed 100% on-device without the cloud and is fully functional with the device in aeroplane mode.

02. Automated Scan

In the absence of cryptographic C2PA signatures, Ape.I analyzes pixel statistics to identify potential pattern variance.

Automated scanning utilizes a custom on-device machine learning model with hardware-acceleration to perform AI analysis.

The resulting confidence score indicates whether the scanned image consists of sub-pixel patterns that are likely to be from a computational AI source, or if they are consistent with a genuine image.

03. Technical Visualization

Ape.I includes a suite of Pro tools that offer further in-depth functionality. These provide additional methods to visualize and uncover artificially created images and edits, including the 16-Grid Deep Scan, Precision Probe, and fully featured Error Level Analysis tools.

Pro tools provide powerful ways to visualize inconsistencies:

Error Level Analysis (ELA)

The Spectral Analysis (ELA) tool visualizes compression artifacts in images. Altered or AI generated photos often exhibit inconsistent noise patterns, lines, and high-contrast edges.

AI models are incapable of directly seeing this metadata, making ELA a valuable forensic tool.

ELA in Ape.I is implemented in two complimentary ways.

Firstly, the automated batch-scan, as well as checking for C2PA and sub-pixel pattern anomalies, also runs an automated ELA scan. If this detects high dynamic range within an image, an alert is displayed with a prompt to check with the Spectral Analysis (advanced ELA tool).

Spectral Analysis features advanced ELA imaging. Providing for fine control and visualization including amplification, visibility, and boost options. The pre-configured baseline allows for incremental adjustments and makes it easy to visually dial in image compression artifacts.

16-Grid Deep Scan

This segments an image into 16 distinct zones for granular scanning. If a zone is selected a recursive sub-grid integrity check will then be carried out.

Precision Probe

Performs localized statistical inference on a specific region of an image. Pixel pattern anomalies in high detail and complex areas are often indicative of AI generated or manipulated images.

Video Content

The scanning engine can also be used to check videos. The system analyzes pixel properties across key frames to return a baseline score.