What does RavenSight moderate?

RavenSight moderates images, video, text, and documents. Visual detection covers nudity, weapons, alcohol, smoking, and drugs. Text and document pipelines extract content (OCR) and flag toxicity — with explainable labels for every decision.

How does the API work?

Upload media via the API (or use the web app), receive a job id, then stream status over SSE (or poll as a fallback) until the run completes. Results include unsafe/safe decisions, labels with confidences, and optional extracted text for OCR jobs. See the developers quickstart for curl examples.

What is Standard, Advanced, and Contextual Detection?

Standard and Advanced are visual detection modes for images and video. Contextual Detection runs visual moderation together with OCR and text toxicity so meme text and overlays are checked in the same run.

How fast is moderation?

Core visual detection often finishes in under a few hundred milliseconds. The API is async: stream job status (SSE) or poll, then fetch the result. OCR, Contextual, and Deep Agentic Review add latency in exchange for richer coverage.

Can I use policies and custom thresholds?

Yes. Create separate media and Docs/PDF (text) policies with their own defaults, or set manual thresholds and labels per run. Use Safe / Review / Unsafe confidence bands and route gray-zone cases to Deep Agentic Review first—with optional human-in-the-loop when you want it.

Do I need human moderators?

Most gray-zone and appeal cases can be handled by Deep Agentic Review—including resolving 98%+ of appeals. Human-in-the-loop remains an option for teams that want final human sign-off on specific decisions.

Who is RavenSight for?

UGC platforms, marketplaces, dating apps, GenAI products, document workflows, and ad creative review — anywhere you need one API for images, video, text, and docs with a reason for every decision.

How do I get started?

Create an account for free credits, generate an API key, and follow the developers quickstart. For enterprise volume or custom needs, contact sales@gotraven.com.

Visual Scan → Multimodal AI ScanGray-Zone → Deep Agentic Review

Detection Queue

SAFE

Clear

Risk 9%

REVIEW

Costume

Risk 58%

REVIEW

Fake blood

Risk 64%

UNSAFE

Weapon

Risk 98%

SAFE

Boxing

Risk 12%

REVIEW

Hooded

Risk 47%

final_decision: safe · mode: contextual

AI-Powered Content ModerationYou Can Trust

Stop harmful content before your users see it, at any scale. One API checks every image, video, and document for unsafe visuals and toxic text, including text hidden in memes. Deep Agentic Review resolves 98%+ of appeals, cutting manual review work for your team.

Images · Video · Text · Docs

Platform

One pass across visual and text signals

Built for UGC & Enterprise — Marketplaces, Dating, GenAI, Docs & Ad Creative. Detect nudity, weapons, alcohol, smoking, and drugs in images & video, plus text toxicity and document OCR — one API, one reason for every decision.

upload_8f2k.jpg

1 request · 6 detectors in parallel

280ms

Nudity
4%
Weapons
97%
Alcohol
12%
Smoking
3%
Drugs
2%
Toxicity (OCR)
91%

final_decision: unsafe · labels: GUN, TOXIC_TEXT

Core visual detection · ~280ms avg

Parallel detectors for nudity, weapons, alcohol, smoking, and drugs—plus OCR and toxicity—so product, trust & safety, and infra share one contract. Opt in to Deep Agentic Review for 98%+ of appeals.

Visual categories

NudityWeaponsAlcoholSmokingDrugs

Signals in one pass

OCR / meme textToxicityMultilingualDeep Agentic Review

Images — JPEG, PNG, WebP, HEIF
Videos — MP4, WebM, MOV, AVI
Docs & text — PDF, DOCX, OCR, toxicity

Policies & Deep Agentic Review

Policy control with Deep Agentic Review for appeals

Separate media and Docs/PDF policies, Safe / Review / Unsafe bands, and gray-zone routing to Deep Agentic Review first—human-in-the-loop when you want it.

Policy Engine & Governance

Separate media and Docs/PDF (text) policies
Safe / Review / Unsafe confidence bands
Gray-zone routing to Deep Agentic Review first; optional human-in-the-loop
Organization-wide defaults & per-run overrides
B2B API Key Authentication
Multi-tenant Isolation
Audit-friendly payloads & org-level usage visibility
Per-key rate limits & quotas aligned to your plan

Deep Agentic Review

Opt-in

Agentic second pass for hard image cases—with an explainable rationale. Designed to resolve 98%+ of appeals without waiting on a human queue. Human-in-the-loop remains available when you want final sign-off.

98%+of appeals resolved by Deep Agentic Review, with an explainable rationale for every decision

Deep Agentic Review first

Route ambiguous gray-zone scores to Deep Agentic Review before a person ever sees them.

Optional human-in-the-loop

Keep a human path when your process needs approve-safe or mark-unsafe sign-off.

Modes & scenarios

Balance speed, accuracy, and context

Choose Standard, Advanced, or Contextual detection—and see how RavenSight handles real moderation scenarios.

StandardFast

Fast parallel detection across nudity, weapons, alcohol, smoking, and drugs

~150ms core detection

AdvancedPrecise

Enhanced accuracy with visual annotations highlighting detected unsafe areas

~280ms core detection

ContextualMultilingual

Highest accuracy with multilingual text detection for global content safety

~350ms+ with OCR path

Detection Scenarios

Example 1: Visual Detection

DETECTION_ENGINE_RESULT: MULTI_REGION_SCAN

How it works: RavenSight runs visual detectors in parallel for nudity, weapons, alcohol, smoking, and drugs. You choose which categories to enable per upload; each returns granular labels (e.g. Belly Exposed, GUN, ALCOHOL) against your moderation policies.

Example 2: Text Analysis

$ analyze --text "This restaurant is absolute garbage and the food tastes like shit. The people running the place are clueless idiots who clearly don't care about customers at all. I regret every minute and dollar wasted on this disaster."

INSULT: 0.96PROFANITY: 0.89TOXICITY: 0.92

How it works: Using our Contextual AI, we analyze the semantic intent. It understands context, enabling it to flag hate speech and profanity even when embedded in professional or conversational text.

Example 3: Contextual Analysis

CONTEXTUAL_ENGINE_RESULT: MULTI_MODAL_SCAN

How it works: This mode combines visual scanning with text extraction. In this example, while the background image is safe, the engine extracts the text "idiots" from the meme. It identifies this as toxic intent directed at a subject, allowing you to flag memes or images that carry harmful messages even when the visual content itself appears benign.

Industries

One API for every surface you moderate

Run RavenSight on the same write path—upload, post, document, or generated output—so policy-breaking media and text never reach your users. Every decision is explainable and stored for audit.

UGC & Trust

Marketplaces, dating apps, communities, and GenAI products

Marketplace & retail photos

Score buyer and seller images before they appear on product pages, reviews, or returns.

Reviews, Q&A & comments

Moderate user-written text pre-publish so harassment and unsafe text never land in public feeds.

Community & profile media

Avatars, covers, and uploads in forums and social surfaces—images and short video on one API.

Dating & matching apps

Check profile photos, chat media, and bios before they reach a match—cut NSFW and catfishing.

GenAI & AI app outputs

Moderate prompts, generated images, and assistant replies before they are shown or stored.

Enterprise & Brand Safety

Document intake, compliance archives, and ad creative pre-flight

Document & PDF intake

Scan PDFs, DOCX, and TXT at ingest—embedded text and visuals—before storage or human review.

Publishing & compliance

Moderate manuscripts, manuals, and knowledge bases with stored, explainable decisions for audit.

Ad creative pre-flight

Block policy-breaking images, video, and copy before spend—not after a network rejection.

UGC & seller ads

Promoted posts and seller creatives on the same write path as organic UGC—one policy, one API.

Workplace & internal uploads

Keep NSFW, violent, and harassing media out of intranets, chat, and collaboration tools.

Features

Enterprise-grade detection

Models trained on diverse datasets for accurate moderation across images, video, documents, and text.

AI Visual Detection

Parallel visual detection for nudity, weapons, alcohol, smoking, and drugs across images and video—with granular labels and confidence scores.

Document & Text Analysis

Extract and analyze text from PDFs, Word documents, and images. Detect toxic language, hate speech, and inappropriate content in text.

Image Text Detection

Advanced text extraction and toxicity analysis from images - perfect for detecting unsafe content in memes, signs, clothing, and screenshots.

Video Frame Sampling

High-speed video analysis with smart frame sampling for live UGC and feed moderation workflows.

Fast core detection

Core visual detectors often finish in under a few hundred milliseconds. The API is async—stream or poll job status, then fetch the result.

Privacy-First Design

Privacy-first design with multi-tenant architecture, built to follow compliance guidelines and keep each customer's data isolated.

Developers

Integrate in minutes, not days

REST API for uploads, status, and results. Documentation covers authentication, core endpoints, and integration patterns—without exposing internal implementation details.

Public integration guides & quickstart
Allowlisted OpenAPI reference (ReDoc)
Signed-in developer portal with live explorer

View documentation API reference

Terminal — bash

1curl-X POST"https://api.gotraven.com/api/v1/upload" \

2-H"X-API-Key: rvn_live_..."\

3-F"file=@photo.jpg"\

4-F"detection_mode=contextual"\

5-F"lang=auto"

6# Response (JSON)

8"run_id":"run_abc123",

9"status":"running"

10}

Pricing

Flexible plans for every scale

Usage-based plans with included monthly units. Pricing isn't listed here—email us and we'll recommend a tier and quote.

Free

Free to start

100 included units / month

30 requests/min · 5 concurrent runs · 100 MB max upload

Content moderation API (images, video & documents)
Analyze Content, API keys & usage overview
Deep Agentic Review for appeals
Standard throughput limits shared across tiers

Startup

Built for pilots & first production

5,000 included units / month

Same rate, concurrency & upload caps as Free

Everything in Free
Higher included volume for production pilots
Email support
Audit-friendly org tools as you scale

Growth

For apps with steady volume

15,000 included units / month

Same rate, concurrency & upload caps as Free

Everything in Startup
Video, custom policies & analytics in-dashboard
Priority support
Room for sustained production traffic

Enterprise

Volume, contracts & compliance

Custom included units & SLAs

Dedicated limits, security reviews & integrations

Custom quotas, contracts & SLAs
Security, procurement & deployment support
Dedicated customer success
Integrations & roadmap alignment

Included units follow your plan configuration in the product. Limits above match the default catalog shipped with RavenSight (Free, Startup, Growth). Developer docs describe how usage is metered.

Built for scale & reliability

Enterprise-grade cloud infrastructure designed to handle your content moderation needs at any scale.

Cloud-First

Fully managed SaaS API

Global Edge

Low latency worldwide

Low latency

Fast core detection

99.9% Uptime

Enterprise SLA

FAQ

Frequently asked questions

Short answers about RavenSight moderation, detection modes, and getting started. API details.

What does RavenSight moderate?: RavenSight moderates images, video, text, and documents. Visual detection covers nudity, weapons, alcohol, smoking, and drugs. Text and document pipelines extract content (OCR) and flag toxicity — with explainable labels for every decision.
How does the API work?: Upload media via the API (or use the web app), receive a job id, then stream status over SSE (or poll as a fallback) until the run completes. Results include unsafe/safe decisions, labels with confidences, and optional extracted text for OCR jobs. See the developers quickstart for curl examples.
What is Standard, Advanced, and Contextual Detection?: Standard and Advanced are visual detection modes for images and video. Contextual Detection runs visual moderation together with OCR and text toxicity so meme text and overlays are checked in the same run.
What is Deep Agentic Review?: Deep Agentic Review is an agentic second pass on hard image cases with a clear rationale. It is designed to resolve 98%+ of appeals without waiting on a human queue. Human-in-the-loop remains available when you want a person to make the final call.
How fast is moderation?: Core visual detection often finishes in under a few hundred milliseconds. The API is async: stream job status (SSE) or poll, then fetch the result. OCR, Contextual, and Deep Agentic Review add latency in exchange for richer coverage.
Can I use policies and custom thresholds?: Yes. Create separate media and Docs/PDF (text) policies with their own defaults, or set manual thresholds and labels per run. Use Safe / Review / Unsafe confidence bands and route gray-zone cases to Deep Agentic Review first—with optional human-in-the-loop when you want it.
Do I need human moderators?: Most gray-zone and appeal cases can be handled by Deep Agentic Review—including resolving 98%+ of appeals. Human-in-the-loop remains an option for teams that want final human sign-off on specific decisions.
Who is RavenSight for?: UGC platforms, marketplaces, dating apps, GenAI products, document workflows, and ad creative review — anywhere you need one API for images, video, text, and docs with a reason for every decision.
How do I get started?: Create an account for free credits, generate an API key, and follow the developers quickstart. For enterprise volume or custom needs, contact sales@gotraven.com.

Ready to protect your platform?

Join companies building safer online spaces with RavenSight AI. Get started today with 100 free credits.