7 October 2026International edition
Vol. I · No.
7 October 2026
AI in Fashion
DAILY
The daily briefing on AI in the fashion business
Where fashion meets artificial intelligence.
Glossary

What is training data in fashion AI?

The examples a model learns from, such as labelled product images, order histories or text.

In short

Training data is the set of examples an AI model learns from, such as labelled product images, order histories or product descriptions. In fashion, the scope, quality and rights status of that data largely determine how well a model performs and whether it can be used safely.

How does it work in practice?

A model only learns what its training data shows it. To build an image tagging model, a team gathers product photos and labels each with attributes such as category, colour and neckline. For a re-order model, the training data might be several seasons of orders per account, enriched with style attributes and delivery dates.

Common sources of training data in fashion include:

  • Packshots and on-model imagery from the product information system.
  • Order, sell-through and return data from ERP and retail systems.
  • Approved product copy, care instructions and size guides.
  • Retailer and customer feedback, reviews and service tickets.

Why does it matter?

If a visual search model was trained mostly on womenswear shot on white backgrounds, it may struggle with menswear photographed outdoors. Gaps in training data turn directly into blind spots in the finished tool. Training data also carries legal weight: brands need to know whether they have the right to use images, models' likenesses and text for training, and publishers increasingly ask the same of AI providers.

How does AI use it?

Every AI model, from a simple forecasting tool to a large language model, depends on training data. Pre-trained models arrive with general knowledge, and brands then add their own data through fine-tuning or retrieval so outputs reflect their products and terminology.

Common pitfalls

  • Unrepresentative samples. Overweighting bestsellers, one region or one body type skews results.
  • Inconsistent labels. If teams describe the same colour or fit differently, the model learns confusion.
  • Unclear rights. Using photography or copy without checking licences and consent creates legal risk.
  • Stale data. Fashion moves quickly, so data from old seasons may teach outdated patterns.

Frequently asked questions

Can fashion brands use their product photos as AI training data?

Often yes, but it depends on the contracts with photographers, models and agencies. Brands should check whether existing licences cover AI training before using the images.

How do you improve training data for a fashion AI model?

Standardise attributes and labels, fill gaps across categories and markets, remove duplicates and errors, and make sure recent seasons are well represented.

All terms

Articles on this term