Skip to main content

Welcome to DeepXL

DeepXL is a forensic AI platform that detects manipulated documents, verifies identities, and extracts structured data — built for lenders, insurers, and government. Integrate via a simple API to catch AI-generated fraud that legacy systems miss.

Our Models

Document Model

Detects AI-generated or altered documents and IDs — bank statements, invoices, passports, driver licenses, and more.

Object Model

Detects whether an image has been AI-generated or manipulated — photos, claim evidence, receipts, and other visual content.

Parsing Model

Extracts and parses structured data from IDs and documents — names, dates, addresses, amounts, and more.

ID-Selfie Model

Verifies identity by matching an ID document against a selfie image or video.

Model Details

Document Model

Detects AI-generated, digitally altered, or tampered documents — including bank statements, pay stubs, invoices, passports, and driver licenses. What it analyzes:
  • Pixel-level manipulation artifacts and metadata inconsistencies
  • Font, layout, and formatting anomalies inconsistent with the issuing source
  • Signs of generative AI fabrication (synthetic documents with no original)

Object Model

Detects whether a photo or visual asset has been AI-generated or manipulated — covering claim evidence photos, product images, receipts, and general visual content. What it analyzes:
  • GAN/diffusion model generation signatures
  • Image splicing, cloning, and compositing artifacts
  • Lighting, shadow, and perspective inconsistencies

Parsing Model

Extracts and normalizes structured data from identity documents and financial records — eliminating the need for manual data entry or custom OCR pipelines. What it analyzes:
  • Identity documents (driver licenses, state IDs, passports, Social Security cards)
  • Financial records (bank statements, pay stubs)
  • Extracts fields such as names, dates, amounts, account numbers, and document numbers

ID-Selfie Model

Verifies that a person presenting an ID document is the same individual captured in a selfie image or short video — combining liveness detection with facial matching. What it analyzes:
  • Biometric similarity between the ID photo and the selfie or video frame
  • Liveness signals to prevent spoofing (printed photos, screen replays, deepfakes)
  • Facial landmark detection and bounding box coordinates for both inputs

How It Works

  1. Authenticate with an API key or Bearer token
  2. Upload a document or image to an analysis endpoint
  3. Receive structured results with fraud scores, quality classifications, and detailed reasoning
  4. Retrieve associated files (originals, heatmaps) via the file retrieval endpoint

Supported File Types

FormatDocument ModelObject ModelParsing ModelID-Selfie Model
JPEG / JPGYesYesYesYes
PNGYesYesYesYes
WebPYesYesYesYes
PDFYesNoYesNo
MP4 / VideoNoNoNoYes

Next Steps

Quickstart

Make your first API call in minutes.

Authentication

Learn how to authenticate your requests.