TurboLens is an OCR tool for image forgery detection and structured document extraction. Its official website is https://www.turbolens.io/。
I. Purpose In workflows such as KYC, lending, insurance, and document processing, TurboLens provides visual forensic determination before images enter OCR and downstream processing, to detect tampered, AI-generated, or AI-edited images, and to perform structured information extraction on documents.
II. Main Capabilities
- Image forgery detection: Perform local tampering analysis on existing images, checking the pixel structure for splicing, copy-move, retouching, and field alterations. Output a genuine or forged determination, the proportion of tampered pixels, coordinates of suspicious areas, and a visualization heatmap.
- AI generation and AI editing detection: Screen whether an image was created by generative AI or modified by generative editing tools, including fully AI-generated documents and AI-edited images. Return a three-state screening result—AI-generated, needs review, possibly human-made—along with an AI probability score.
- The two layers of detection are independent of each other: image forgery detection targets local tampering in existing images, while AI image detection assesses generative origin. The two results are returned separately, making it easy to route them into the corresponding review process according to their respective risks.
- Specialized OCR and document intelligence: After authenticity screening is completed, documents can be classified, structured fields can be extracted, and complex Southeast Asian layouts and scripts can be handled. Specifically, this includes: invoice and receipt OCR (extracting vendor, tax, line item, and payment fields), identity and passport OCR (extracting structured identity fields from Southeast Asian national ID cards, passports, and multilingual scans), contracts and legal documents (converting to structured clauses, parties, dates, and reviewable fields), and regional forms and languages (handling tax forms, customs documents, and mixed-language layouts).
III. How to Use Before OCR and downstream processing, submit document pages or images to the API (Step 1); run pixel-level forgery analysis and, when enabled, synthetic media screening (Step 2); clean images automatically continue through the flow, while flagged results enter a review workflow (Step 3). Teams can control thresholds themselves and decide whether flagged images are rejected, escalated, or placed in a manual review queue. The official description states that authenticity checks can be added without rebuilding existing pipelines, retaining existing OCR, face matching, watchlist screening, and case management tools, and adding authenticity signals through REST endpoints. It returns structured JSON responses containing pixel-level evidence, configurable review routing, and support for batch processing workflows.
IV. Applicable Scenarios
- KYC and account opening: Block tampered identity information before account creation, and screen passports, national ID cards, driver's licenses, and supporting documents for photo replacement, textfield editing, and synthetic generation.
- Lending: Before credit decisions, detect suspicious edits in payslips, bank statements, tax forms, and other proof of income.
- Insurance: Before claim review, check receipts, invoices, medical documents, and claim photos for local tampering or generative editing.
V. Limitations and Notes The synthetic media detection service is currently in an early experience stage. It is used to screen fully AI-generated images and images modified by generative editing tools, and its results are designed to assist rather than replace manual review and business rules. Visual forensics is generally recommended to be placed after upload and before OCR/extraction/approval. Pricing information is not yet available.
