Training data (AI training data)
Japanese 学習データ（AIの学習データ）
Training data is the text, images, video and other material an AI model reads while it is being built. What a model learned from shapes both the quality of its output and the rights questions around it.
An AI model does not know the answers from the start. It learns by reading many examples and slowly adjusting the numbers inside it (its parameters). The set of examples it reads is the training data. Just as a cook's style depends on which cookbooks they learned from, what an AI is good at and where it slips up depends on what it read. Collected material is usually split into a part for training and a part for checking performance, and the checking part is kept out of training.
Watch out Whether the training data included other people's work becomes a rights question when you use AI output in your business. Many tools do not disclose the details of their training data, and the OSI's Open Source AI Definition requires information about the training data to be shared. Before using AI-made images or video commercially, read the tool's terms and what it says about its training data, and check that the result does not look like an existing work.
Source: Machine Learning Glossary · The Open Source AI Definition – 1.0 · AI Content on Steam · ToonCrafter: Generative Cartoon Interpolation (checked 2026-10-26)