Sightengine
Sightengine
Content moderation and AI-content detection APIs
Sightengine provides APIs that automatically assess, filter and moderate images, videos and text. Its visual moderation spans 150 classes covering nudity, hate, violence, drugs, weapons and self-harm, with asynchronous and live-stream video handling on higher tiers. Alongside moderation it detects AI-generated images, video, speech and music, identifies deepfakes and face swaps, finds near-duplicates and spam through custom disallow lists, extracts text with OCR, validates profile pictures and estimates age groups.
Primary Functions
image and video moderation
text moderation
AI-generated content and deepfake detection
near-duplicate and spam detection
OCR
Features
- Image and video moderation across 150 classes including nudity, hate, violence, drugs, weapons and self-harm
- Advanced video and live-stream moderation with configurable analysis frame rates
- AI image, AI video, AI speech and AI music detection, plus deepfake and face-swap identification
- Text moderation for sexual, hateful, violent and insulting content, personal information, links and usernames
- Near-duplicate and image-spam detection with custom image and video disallow lists
- OCR and QR-code moderation to catch text and codes embedded in images and video
- Profile-image validation and age-group estimation
- Dashboard workflows supporting both automated and hybrid (human-in-the-loop) moderation
Limitations
- Free and Starter have no live-stream support and cap videos at 50MB; Starter allows only one concurrent video job
- Throughput is limited to 1 request per second on both Free and Starter, reaching 10 req/s only on Pro
- Audio moderation, image and video custom search, near-duplicate detection, liveness and recapture detection, age estimation and AI music detection all start at the Pro tier
- Custom models and classes, geo-fencing, SAML SSO, regional data processing, enterprise SLAs and a dedicated support engineer are Enterprise-only
- Vendor pages give different image-moderation class counts (150 on the home page, 120+ in the products menu) and publish no accuracy or benchmark figures as text