Skip to content
Hive logo

Hive is a leading provider of cloud-hosted pre-trained AI models across a wide range of use cases. Today, hundreds of companies are passing billions of pieces of content through Hive APIs each month to understand image, video, text, and audio data.

HIVE

Visual Moderation API

Hive’s Content Moderation API contains over 40 classes that will process images and video in near real-time. Our model classes include things like nudity, violence, hate imagery and others. This will allow users to ingest metadata quickly and perform any downstream removal of content deemed innapropriate. The overall moderation taxonomy is structured in a way that gives detailed granularity allowing for more nuanced moderation. Customers can decide which classes to use as the most relevant signal for content removal.

Extensions Integration Overview:

Hive’s visual moderation API integration is quite simple to set up – we provide several options for the submission of images or videos into our REST API. Hive supports both synchronous and asynchronous API interface protocols. The asynchronous endpoint is preferred for users who are submitting their volume in large batches users submitting tasks containing large files (i.e. longer videos or audio clips).

The asynchronous endpoint immediately sends a response acknowledging receipt of the task, along with a unique ‘task_id’. It then closes the connection. Once the task is completed, Hive will send a POST request to the provided callback_url containing the completed task’s results. Customers simply need to set up a callback server, and Hive’s API will post callback responses to the customer specified callback_url described above.

Our visual moderation model classifies an entire image into different categories and assigns a confidence score for each class – this is important to note as each image or frame of video recieves several scores.

Extension Features Overview:

Hive Visual Moderation Extension Features:

  1. Over 40 classes
  2. 500ms response times (Latency optimized for real-time processing)
  3. 99% precision at 90% recall (best in class model accuracy)
  4. Custom Classes Support
  5. API updates quarterly
  6. Support for live-streaming technologies (RTMP,HLS)