Insights

AI tagging: what it structures, what it does not decide.

AI tagging describes what is in an image. It does not decide what the image is for. Accuracy, how many tags are right and how many wrong, is the smaller question. A wrong tag makes an asset harder to find. A recorded decision nobody made changes what a team believes was agreed.

A mountain biker descends a steep scree slope of pale gravel, with a broad valley and layered ridges under a heavy, light-shot sky behind. Photographed by Martin Bissig in Patagonia.

Photo: Martin Bissig

Photo: Martin Bissig

AI tagging produces description, not judgment

AI tagging is a model reading attributes from the pixels: objects, scenes, colors, text in the image, whether faces are present. Anyone can check them by looking at the asset, and at volume they are worth having, because a library growing by thousands of assets a year outpaces manual keywording. What a model cannot read is the record of agreements around the image: which campaign it belongs to, whether it was approved, what the license permits. It can report that a frame shows a building at dusk, not that the image is licensed for one region and one year.

Accuracy errors cost a search, authority errors cost a commitment

Models do well on recognizable objects and legible text and less well on abstract work and small brand marks, and that shifts with every model generation. A wrong tag costs one search and is fixed by correcting the metadata. A system that marks an asset as cleared when nobody made that call sends the claim out of the library, and the question arrives after publication, when correcting the metadata no longer helps. One is a retrieval cost. The other is a commitment made on someone's behalf.

Better models do not change what a decision is

A model writes attributes. People write states: what was approved, what stage the asset has reached, who may reach it. An approval is not a fact waiting to be detected. It is created by a person with the standing to create it, and the same holds for access and for usage rights. Accuracy never converts description into decision, however far accuracy goes.

The moodcase Library detail view of the Patagonia project. The photograph fills the frame while the Info panel lists machine-suggested AI Tags including Slope, Bicycle, Gravel and Cycling, above photographer and copyright, with the human-written Alt Text and Keywords fields still empty.

What moodcase does with this

In moodcase Visual Asset Management, AI Tagging suggests tags as assets enter the library, Asset Metadata holds the structured fields, and Global Search returns assets by attribute across the workspace. Workflow Statuses are set by the people responsible for them.

The design decision is to keep the two apart. The tagging can keep improving, and what the library records about approval and access does not move with it.

Where this fits

When a library is too large to describe by hand and people act on what it says, the line between a suggested tag and a recorded decision starts to matter. In moodcase tags arrive at upload, and a state changes only when someone sets it.

Visual assets need more than a folder. See how moodcase handles the full workflow.

Try all features for 7 days. No credit card required.

AI Tagging

Metadata

Search

Creative Teams