Enhancing Content Creation via Gen AI

Automate a time-consuming work process that delivers low-quality results. Collecting images requires massive efforts from the client's team, involving tasks such as selecting, formatting, and labeling large volumes of visual data. This process was long, disorganized, and required extensive man-hours. Moreover, it often results in low work efficiency, with the quality of the resulting picture sets falling short of desired standards. All visual materials were stored on team members' personal devices, affecting workflow efficiency.* The client had an idea to introduce a "Look-A-Like" feature, which would offer admins the opportunity to make photo selections similar to specific images.

*With extensive data scraping and Data Engineering expertise, Dataforest has developed a simple and efficient Gen AI solution that automates the required workflow and delivers faster and better-quality results.

Using data scraping and an advanced image recognition model, LLaVA, the solution automates detecting, analyzing, labeling, storing, and retrieving images. Therefore, it eliminates human involvement and review, saving significant person-hours. It also delivers results that impressively exceed the quality of human work by having a tailored labeling system for 20 attributes and reaching 96% model accuracy.* The developed solution labels data and stores it in a unified database. It displays imagery data in a Gallery with a friendly UI, where a filter system allows users to select images based on specific criteria.* For the "Look-a-Like" feature, Dataforest developed a solution using a vector database. The application retrieves similar images and runs additional matching based on respective labels and predefined characteristics.

Dataforest developed an easy-to-use Gen AI solution for image collection that automates all work within one UI-friendly workspace, saving time & manpower and delivering better-quality results.

‍

The solution uses LLaVA, an advanced image recognition model which helps to label 20+ attributes. It delivers high-quality results in the image set selection processing 3250 images per 1 hour and achieving 98% model accuracy. All images are stored in a unified, automatically updated repository and can be retrieved using filters via user-friendly UI.

‍

By leveraging our solution, the client revolutionized his workflow, saving human hours and getting faster and higher-quality results. Moreover, this web solution can be used for the new functionalities developed to work with his data.