AI IMAGE MODERATION · v3.2

AI-Powered
Image Moderation

High-precision visual risk detection at scale. A granular taxonomy with 976 detection labels across 8 risk categories — built to keep global platforms brand-safe and compliant.

Start Free Trial Get in Touch API Documentation

REQUESTS / SEC

12.4↑ 8.2%

FLAGGED TODAY

3.2M

AVG LATENCY

142ms

image-moderation · live

ANALYZING

DETECTIONS 4 / 976

porn.female_nipple

874

prohibited.firearm

723

qr_code.wechat

645

minor.teenager

412

BLOCK policy.csam_strict

processed in 142ms

8

Risk Categories

164

Sub-categories

976

Detection Labels

99.2%

Recall on Core

Label Architecture

Three-tier taxonomy.
Down to the last detail.

Every flagged image returns a structured label path — from broad category down to specific sub-type. Configure policies at any level of granularity.

moderation_result.json

L1 Regulated & Prohibited Content

Top-level moderation domain

L2 Illicit Drugs

Sub-category within the L1 domain

L3 Cannabis Plant

Specific detector returned by the API

Block, review, allow — at any granularity

Set platform-wide rules at L1 to block all sexually explicit content. Or get surgical at L3: allow artistic nudity through, but flag AI-generated nudity for review. Combine with confidence thresholds (0–1000) to route ambiguous cases to human moderators.

L3 labels are powered by over 3,000 underlying L4 detection variants — covering specific personas, named objects, and stylistic patterns — exposed through a clean L3 API surface optimized for policy configuration.

Every detector is independently tunable. Override defaults per workspace, per content channel, or per content type.

Detection Coverage

Eight categories.
Every shade of risk.

Each category breaks down into focused sub-categories and specific detectors. Click any module to drill into the complete label reference.

⬢

Sexually Explicit Content

色情

CRITICAL

310

L3 labels

TOP SUB-CATEGORIES

Sexual Acts 23

Female Nudity 21

Explicit Female Nudity 19

Explicit Female Nudity – Illustrated 17

Animal Mating Behavior 11

Explicit Male Nudity 10

SAMPLE DETECTORS

AI-Generated Anomalous Male ChestAI-Generated Abnormally Large BreastsCat Genital ExposureAvian Mating BehaviorAnimal Genital ExposurePixelated/Censored Sexual Content

⬡

Suggestive Content

性感

MODERATE

72

L3 labels

TOP SUB-CATEGORIES

General Nudity 31

Female Suggestive Content 12

Suggestive Poses & Actions 11

Intimate Affection 9

Male Suggestive Content 4

Sexual Provocation & Innuendo 3

SAMPLE DETECTORS

Female Back ExposureButtocks Close-UpFemale-Female Impending KissMale Leg & Foot Close-UpMinor Female in Suggestive ClothingBody-Tight Clothing

◆

Regulated & Prohibited Content

违禁

CRITICAL

126

L3 labels

TOP SUB-CATEGORIES

Prohibited Items 22

Prohibited Behavior 18

Controlled Bladed Weapons 12

Illicit Drugs 11

Gambling Equipment 8

Military Weapons 7

SAMPLE DETECTORS

Adult Pregnancy – IllustratedAlcohol BottleBody PiercingsBlade / RazorDrug Use ParaphernaliaDrink Spiking

◈

Violence & Extremism

暴恐

CRITICAL

36

L3 labels

TOP SUB-CATEGORIES

Violence & Extremism Scenes 22

Graphic Gore Scenes 8

Terrorist Organizations 3

Violent & Extremist Media 2

Violence & Extremism (General) 1

SAMPLE DETECTORS

Animal GoreTerrorist Organization MembersViolence & Extremism (General)Animal Attacking HumanSubverted Animation & IP Characters

⬟

Spam & Advertising Violations

广告

MODERATE

105

L3 labels

TOP SUB-CATEGORIES

App Screenshot Ads 13

Contact Information Solicitation 13

Deceptive Click-Bait Ads 13

General Industry Ad Creatives 10

Advertising Scene Contexts 9

Low-Quality Ad Creatives 6

SAMPLE DETECTORS

Light-and-Shadow Text AdAdult-Content Account SolicitationChat Screenshot AdCalculator Screenshot AdContact Account SolicitationAnnotation Graphic Overlay

⬣

Minor Detection

未成年人

CRITICAL

8

L3 labels

TOP SUB-CATEGORIES

Child Sexual Abuse Material (CSAM) 2

Adults 1

Child Subjects 1

Child-Targeted Violence 1

Infants 1

Preschool-Age Children 1

SAMPLE DETECTORS

Adult SubjectChild Nudity (CSAM)Child Subject PresentElsagate / Disturbing Child-Targeted ContentInfant SubjectPreschool-Age Child Subject

⬢

Politically Sensitive Content

涉政

HIGH

302

L3 labels

TOP SUB-CATEGORIES

Domestic Political Symbols 65

Anti-Government & Separatist Movements 31

International Political Leaders 30

Sensitive Political Events 22

Political Leaders (General) 17

Core Historical Leaders 16

SAMPLE DETECTORS

COVID-19 PandemicAnti-China FiguresCore Historical Leader – CaricatureChinese Yuan (RMB)Industrial AccidentForeign Disreputable Figures

▦

Embedded Code Detection

二维码

LOW

17

L3 labels

TOP SUB-CATEGORIES

In-App QR Code 2

Physical-Goods QR & Barcode 2

Virtual-Goods QR Code 2

Alipay QR Code 1

Background-Embedded QR Code 1

Linear Barcode 1

SAMPLE DETECTORS

Alipay QR CodeBackground-Embedded WeChat QR CodeWeChat Mini-Program Standard QR CodeLinear BarcodeMini-Program QR CodeOther QR Codes

Why DeepCleer

Why Trust & Safety teams
switch to DeepCleer.

Granular & Industry-Tailored Taxonomy

Access highly precise, comprehensive, and personalized moderation logic that fits your specific business needs. 976 labels with rule profiles for AIGC, dating, social, ad-tech, gaming, and marketplace.

Account-Level Intelligence

Identify and flag high-frequency offenders at the account level — not just individual content. Enables proactive intervention and robust governance against repeat policy violators.

Global-Scale Elasticity

Seamlessly handle billions of inspection requests daily with zero-latency protection. Multi-region deployment across the US, EU, and APAC — your traffic volume never becomes a constraint.

Agile Intelligence & Rapid Iteration

Models feature hourly updates — your defenses evolve constantly to counter the latest bypass techniques, emerging meme patterns, and AIGC adversarial inputs as soon as they appear.

Core Features

Built for production.
Engineered for scale.

[01]

Dynamic Policy Engine for Rapid Adaptation

Combine our granular 976-label taxonomy with an intuitive configuration interface. Toggle detectors, set custom thresholds per L1/L2/L3, and pivot instantly to meet emerging regulatory requirements.

Operators ship policy changes in seconds, not deployment cycles — critical when crisis content emerges or new compliance rules take effect.

porn.female_nudity threshold: 500

violence.severe_gore threshold: 400

minor.csam threshold: 200

drugs.cannabis threshold: 700

ad.deceptive_click threshold: 600

qr_code.wechat threshold: 800

suggestive.lingerie threshold: 650

[02]

Proprietary Data Lakes Powering Continuous Iteration

Our models are trained on a massive, ever-expanding repository of image samples. We maintain specialized asset libraries — covering celebrities, animals, objects, and complex scenarios — ensuring our AI remains battle-tested against the latest visual trends.

Bypass-resistant by design: every adversarial pattern flagged in production feeds back into training within hours.

celeb_db · 1.2M

aigc_db · 4.6M

object_db · 890K

scene_db · 2.1M

[03]

Ensemble Architecture for Industry-Leading Precision

We achieve 99%+ accuracy by synthesizing a diverse stack of State-of-the-Art (SOTA) deep learning architectures — including Inception, ResNet, MTCNN, InsightFace, EAST, and CRNN.

This multi-dimensional ensemble approach ensures precise identification of even the most subtle visual violations — from heavily-pixelated nudity to QR codes embedded in background textures.

Inception 98.4

ResNet 97.6

MTCNN 99.1

InsightFace 98.2

EAST + CRNN 96.8

ENSEMBLE OUTPUT
99.2%

Use Cases

Image recognition
across every scenario.

From dating apps to ad networks — Trust & Safety teams across industries use DeepCleer to enforce content policies at scale.

SOCIAL · UGC

User-generated content platforms

Pre-screen profile photos, posts, and DM attachments. Configurable per content type — strict for public posts, lighter for private messages.

nudityviolencesuggestiveminor_detect

DATING · MATCH

Dating & matchmaking apps

Verify profile authenticity and enforce strict no-nudity policies. Age-group detection prevents inappropriate matching across age boundaries.

profile_validatesuggestiveage_estimate

AD-TECH

Ad networks & ad platforms

Detect creatives with deceptive UI, fake close buttons, low-quality artifacts, or industry-compliance violations before they reach publishers.

deceptive_clickindustry_violationwatermark

E-COMMERCE

E-commerce marketplaces

Block listings featuring prohibited items — weapons, drugs, wildlife products, regulated pharmaceuticals — before they appear in your catalog.

prohibited_itemindustry_regqr_code

GAMING · LIVE

Gaming & live-streaming platforms

Moderate streamer thumbnails, in-game screenshots, and chat-attached images. Detect game-boosting and account-poaching advertisements.

live_screenshotgame_boost_adsuggestive

AIGC · GENAI

AI-generated content platforms

Filter outputs from text-to-image models. Detect AI-generated nudity, anatomical anomalies, and policy-violating prompts — pre and post generation.

ai_nudityanatomical_anomalyauthenticity

COMMUNITY · FORUM

Communities & forums

Keep discussion spaces civil. Detect graphic violence, hateful symbols, doxxing imagery, and off-platform traffic-diversion content.

violence_terrorhate_symboltraffic_divert

EDUCATION · KIDS

Education & children's platforms

Multi-layer CSAM detection across real, illustrated, and AI-generated content. Strict age-group thresholds with NCMEC-aligned reporting hooks.

csamchild_violenceage_detect

Developer Experience

Drop-in API.
Sensible defaults.

A single endpoint returns structured JSON with the full label hierarchy and confidence scores. SDKs available for Python, Node, Go, and Java.

# Moderate an image against the full label set
curl -X POST https://api.deepcleer.com/v1/image/check \
  -H "Authorization: Bearer $DC_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "image_url": "https://your-cdn.com/photo.jpg",
    "labels": ["all"],
    "thresholds": { "porn": 500, "violence_terror": 600 }
  }'

from deepcleer import Client

client = Client(api_key=os.getenv("DC_API_KEY"))

result = client.image.check(
    image_url="https://your-cdn.com/photo.jpg",
    labels=["porn", "violence_terror", "minor"],
    thresholds={"porn": 500, "violence_terror": 600}
)

if result.decision == "block":
    flag_for_removal(result.labels)
elif result.decision == "review":
    send_to_queue(result)

{
  "decision": "block",
  "labels": [
    {
      "l1": "porn",
      "l1_name": "Sexually Explicit Content",
      "l2": "female_nudity",
      "l2_name": "Female Nudity",
      "l3": "nvxiongloudian",
      "l3_name": "Female Nipple Exposure",
      "confidence": 874,
      "bbox": [120, 45, 280, 210]
    }
  ],
  "processing_time_ms": 142
}

Ready to secure your platform?

Get a personalized demo with your content types and use cases. Start with 10,000 free moderation requests — no credit card required.

Request a Demo Talk to Our Expert