Hive
Hive
Usage-priced APIs for content moderation and content understanding
Hive provides pre-trained AI models, delivered as REST APIs, that classify, search and generate content at scale. The moderation line covers Visual Moderation, Text Moderation, Audio Moderation, OCR Moderation and CSAM Detection, alongside AI-generated content and deepfake detection, logo and celebrity recognition, OCR, contextual scene classification, translation, speech-to-text, media search and an AutoML platform for custom models. The API returns classification metadata only, confidence scores for visual models and severity levels for text, which customers apply against their own policies; Hive itself does not remove content or ban users. A self-serve V3 API with default rate limits sits alongside enterprise V2 projects and a Moderation Dashboard for review and escalation.
Primary Functions
Features
- Visual, Text, Audio and OCR moderation returning per-class confidence scores for visual models and 1-3 severity levels for text
- CSAM Detection via a classifier API, a hash-matching API and a combined API, plus a CSE text classifier, built with Thorn's specialized model
- AI-generated content and deepfake detection across image, video, audio and music
- Live stream moderation over RTMP and HLS, with audio segment length configurable from 2 to 60 seconds and video sampled at a default 1 frame per second
- Synchronous and asynchronous submission with published latency benchmarks (single image frame ~250ms p50; 1 hour of video ~3 minutes p50)
- AutoML for training custom models on your own datasets, plus the Hive Moderation Dashboard for fine-grained content review and escalations
- A wider model catalogue: OCR, logo and location recognition, IP likeness and celebrity recognition, face detection and similarity, people counting, media search, translation, speech-to-text and generative text/image/audio/video
- Feedback channel through the API to report false positives and false negatives against a task ID
Limitations
- Self-serve V3 rate limits are low by default - 100 requests/day for Visual Moderation and Text Moderation - and sustained higher volume requires an Enterprise upgrade arranged through sales
- Text moderation tasks are capped at 1024 characters and Hive will not raise the limit; longer content must be split into segments and submitted as separate tasks
- The APIs do not support batch processing; batch submission is only through the dashboard UI (under 100 local files, or a CSV of hosted URLs for 100-50,000 tasks)
- Much of the catalogue is quote-only: CSAM Detection, Demographic Attributes, GARM brand-safety variants, video processing for Visual Moderation and every 'higher limits' tier are Contact Sales
- There is no genuinely free tier - the '$50+ in Free Credits' require a payment method on file - and customer data is retained for 14 days by default unless you request a shorter period