Spawning AI

Spawning AI offers the Data Diligence package, an API and open-source Python library that automates the checking of media URLs against opt-out registries so AI developers can filter out creator-restricted content before training their models.

The Data Diligence package allows AI model trainers to automatically check each media item they intend to download against Spawning's Do Not Train Registry and other machine-readable rights reservations (such as HTTP headers and ai.txt). Items registered in the DNTR are not downloaded, and items with other rights reservation signals are excluded from the training dataset. Spawning describes the package as supporting compliance with the EU's text and data mining copyright exceptions and the upcoming AI Act. The package aggregates multiple opt-out standards so developers do not need to manage each registry separately.

Spawning AI offers Source.Plus, a marketplace that uses classifier models to detect nudity, gore, and personally identifiable information in images, helping AI developers curate consenting, high-quality training datasets.

Source.Plus is seeded with nearly 40 million public domain and Creative Commons Zero (CC0) images sourced from libraries and museums worldwide. To help developers filter the dataset, Spawning trained classifier models designed to detect nudity, gore, and personally identifiable information within the image collection. The platform also filters out images whose creators have opted out via the Do Not Train Registry. Users can adjust the detection thresholds of these classifiers to flexibly filter the dataset to their needs. Source.Plus launched in limited beta in 2024, with plans to expand to licensed content from rights holders.

Spawning AI offers Kudurru, a network-based tool that detects web scrapers collecting images for AI training and allows website owners to block them or send back substitute images.

Kudurru works by having member websites analyze incoming web traffic to identify and blacklist IP addresses associated with AI data scrapers. Website owners can choose to block detected scrapers outright or optionally serve them alternate images (a technique sometimes called 'data poisoning'), which can corrupt the data a scraper collects. Sites enrolled in Kudurru are also automatically added to Spawning's Do Not Train Registry. As of November 2023, Kudurru was in beta and available as a WordPress plugin blocking over 400 identified scraper bots.