Quick heads-up for anyone using Natural Language Search (NLS) in the Asset Library: You’ll get the best results by toggling between NLS and regular search depending on what you’re looking for.
Here’s why:
Regular Search
Use this when you’re searching for something specific, like a SKU, file name, or tag.
Our standard search and NLS actually use different query methods/backends, so structured metadata (like a product number) is matched more reliably with regular search.
Natural Language Search (NLS)
Turn this on when you want to search by description or concept, like:
- “men wearing hat”
- “winter office scene”
- “modern healthcare workspace”
For Natural Language Search (NLS), we use a multi-modal embedding model to retrieve results. This type of machine learning model can translate both images and text into multi-dimensional numerical vectors. Each dimension represents a feature the model has learned during training. Now, replicate that across hundreds of dimensions and you get a model capable of capturing complex aspects of reality in numerical form, across both text and images. When you run a natural language query, the search engine converts your query into a vector and then looks for images in the asset bank with