Clean Data

Clean Data builds clean models

We train our in-house models on purpose-built Clean Data collections of images and other creative work. We call the resulting models “clean” because we are building them so their training composition and contributing works can be accounted for—not treated as an unknown pile of files.

Keep the source attached

For works in a training collection, we want to retain useful provenance: creator, source, applicable license, and the collection or role the work was used for. That context should remain available to the model workflow and the people using it, instead of disappearing when files are gathered together.

A separate, open asset library

Our shared asset library is made up almost exclusively of CC0 and CC BY content. We keep the applicable source and license information visible and credit creators when required by the specific license version.

The asset library and model-training collections are separate parts of the system. People can also add their own assets to their library; a user upload is not automatically CinterForge-sourced or CC-licensed content.

Support the people who make the culture

Our aim is to support creators, collectors, and users by making it easier to understand where content came from and how it can be reused. We want our models and tools to account for how training collections were composed and which creators contributed source material.

Explore the Cards and portable workflows that help describe creative work, or browse our content and downloads and their license information.