Smarter, Faster, Automated Image Organization

Nyckel's real-time API for custom machine learning is a game-changing to for digital asset management. Here is why.

Oscar Beijbom

Nov 2022

The challenge

Customers turn to digital asset management software (DAM) to save time and organize overwhelming libraries of images. But managing the management software can drain resources: employees are forced to spend valuable time manually applying tags to images so that they can find them again later.

Expectations vs reality

Customers expect intelligence and speed from modern software. No one wants to manually label thousands of images, and customers expect some smart help from their DAM software. But the current marketplace of “smart” tagging is limited.

Let’s take a look at the landscape of automation products available to DAM companies:

Generic AI-auto-tagging

Services like AWS Rekognition and Google Vision API rely on generic, canned models that have been trained to recognize a suite of everyday objects but are not customized to the customers needs. They can recognize an apple, but they can’t say whether it’s Fuji or Gravenstein.

Custom AI-auto-tagging

Services like AWS Rekognition Custom and Google Vertex AI allow users to train custom models: after labeling a few photos of apples as Fuji, and others as Gravenstein, the model can distinguish between these apple breeds.

But there’s a catch: after tagging the apple photos, users have to tell the model to train and then wait an average of two to five hours while it does so. Then, the user needs to remember to deploy the new model. Every time the model is refined - maybe the user notices some apples are mislabeled and fixes it – it needs to be retrained for the full time and then actively redeployed.

The faster, smarter solution

Nyckel’s fast, agile API trains and deploys in seconds, achieving the same or better accuracy as AWS Rekognition and Google Vertex AI. It automatically re-trains and re-deploys every time data labels are refined or corrected. This closes the loop with the user, giving them instant reward for their labeling work – keeping them engaged in refining the model, and providing a seamless, truly automated experience.