Detect text in images with OCR moderation

Text moderation reads the text fields users type into your platform. OCR moderation reads the text embedded inside images, which is the channel users use to bypass typed-text filters. Lasso runs both through the same four-layer pipeline, so a phone number in a photo is treated the same as one in a chat message.

It runs in four layers. Lasso's OCR first extracts text from the image across 200+ languages, then your custom rules filter anything your platform does not allow. The AI Moderator weighs context in captions and overlays and routes each item to auto-remove, flag, or approve. Anything in a grey area queues for human review with the extracted text, original image, AI reasoning, and a suggested action.

Yes. Lasso extracts rotated, colorized, tiny, and stylized text, including meme overlays and obfuscated leet speak. It also detects QR codes that redirect to external sites and reads payment addresses embedded in images. Your custom models define what counts as a violation.

Yes. Lasso reads phone numbers, social handles, Telegram usernames, and payment addresses hidden inside profile photos, listings, and avatars. Once extracted, that text runs through your rules so off-platform solicitation is flagged before it reaches your users. You decide per platform which contact patterns to block.

99.9% detection accuracy on policy violations. 99.9% text extraction accuracy across 200+ languages.

Yes. Confidence scores range 0–100. Set your automation threshold per policy: high-confidence flags auto-action, borderline flags route to human review. You tune it per policy and per industry.

Lasso's OCR extracts text in 200+ languages and scripts. CJK, Arabic, and Devanagari are individually optimized. You can configure different moderation rules per language if needed.

Yes. Common techniques detected: leet speak (c0d3 for code), homoglyph substitution (Cyrillic vs Latin lookalikes), strikethrough formatting, tiny text, QR codes redirecting to external sites, payment addresses in images. Your custom models define what counts as evasion.

Text extraction in realtime, no delay experienced by users. At millions of images, Lasso maintains predictable latency for real-time moderation.