Visual Moderation - Overview

DeepCleer Visual Moderation is an enterprise-grade content moderation system designed to detect visual risks across photographic, illustrated, and AI-generated images at scale. Built on a granular three-tier label taxonomy with 976 detection labels across 8 risk categories, it empowers Trust & Safety teams to enforce content policies with bounding-box precision.


What Visual Moderation Detects

Visual Moderation covers eight top-level risk categories, each broken down into focused sub-categories and specific detectors. Every flagged image returns a structured label path — from broad category down to specific sub-type — enabling policy configuration at any level of granularity.

CategoryDescription
Sexually Explicit ContentExplicit nudity, sexual acts, adult products, and pornographic illustrations across real photographs, AI-generated images, and artistic works. Includes Child Sexual Abuse Material (CSAM) detection that triggers mandatory reporting workflows.
Suggestive ContentSuggestive but non-explicit content including revealing clothing, intimate poses, and provocative expressions. Suitable for age-gating and platform tone control.
Regulated & Prohibited ContentRegulated items and prohibited behaviors including illicit drugs, weapons, gambling equipment, tobacco, alcohol, personal identification documents, military weapons, tattoos, and minor-protection violations.
Violence & ExtremismViolent and extremist content including terrorist organization symbols and members, graphic gore (graded mild to severe), self-harm, suicide, executions, and other disturbing imagery.
Spam & Advertising ViolationsUnauthorized or non-compliant advertising including deceptive click-bait UI, app screenshot ads, watermarks, low-quality creatives, contact-information solicitation, and industry-specific advertising violations.
Minor DetectionAge classification system distinguishing infants, preschool children, teenagers, and adults. Supports CSAM detection and minor-specific policy enforcement.
Politically Sensitive ContentPolitically sensitive content including political leaders, separatist symbols, terrorist organization iconography, fascist imagery, religious cult content, hate symbols, and sensitive historical events. Configurable by regional jurisdiction.
Embedded Code DetectionQR codes, barcodes, and other machine-readable codes embedded in images — distinguishing platform-specific codes, physical product codes, and traffic-diversion codes.

Three-Tier Label Architecture

Visual Moderation organizes its detection capabilities into a structured hierarchy. Every flagged image returns this complete path, enabling precise policy enforcement at any level.

TierRoleExample
L1 — CategoryTop-level moderation domainRegulated & Prohibited Content
L2 — Sub-categoryDomain area within the L1 categoryIllicit Drugs
L3 — DetectorSpecific label returned by detectionCannabis Plant

Set platform-wide rules at the L1 level to block entire categories. Or get surgical at L3: allow artistic nudity while flagging AI-generated nudity for review. Combine with confidence thresholds to route ambiguous cases to human moderators.


Coverage at a Glance

ItemSpecification
Top-level Categories (L1)8
Sub-categories (L2)164
Detection Labels (L3)976
Supported Image FormatsJPEG, PNG, WebP, GIF, BMP, TIFF, HEIC
Bounding Box OutputSupported for spatial-specific detectors
Multi-label OutputMultiple labels returned per image

Designed for Production

Configurable at every layer. Set independent confidence thresholds per L1, L2, or L3 label. Block, review, or allow at any granularity. Override defaults per workspace, per content channel, or per content type.

Jurisdiction-aware. Pre-built policy profiles for North America, EU, Middle East, and Southeast Asia. The Politically Sensitive Content category is fully configurable by regional jurisdiction.

Built for AIGC. Dedicated detectors for AI-generated nudity and anatomical anomalies — purpose-built for AI-generated content moderation workflows.

Child safety first. Multi-layer CSAM detection across real, illustrated, and AI-generated content with mandatory-reporting workflow integration.


How to Use This Documentation

Each category sub-page provides a complete reference for the labels within that L1, including:

  • Overview — Category metadata, scope, and typical use cases
  • Sub-categories — Summary table of all L2 sub-categories with label counts
  • Label Definitions — Complete L3 label list with detection definitions

For integration details, request/response schemas, error codes, and code examples, refer to the Visual Moderation API documentation under the API section.