Labelbox has introduced enhanced video search capabilities within its Catalog, powered by Google’s Gemini Pro Vision, enabling natural language search across video datasets. This integration allows users to search and group similar videos using embeddings to facilitate data curation and identify outliers. By leveraging natural language, data teams can quickly locate specific events in large video collections and create classifiers to highlight similar occurrences in real-time streams, thus improving the retraining of task-specific models. The new cluster view feature enhances this process by identifying groups of similar videos, thereby simplifying outlier detection and aiding in the labeling of data to refine model performance. This development is part of Labelbox’s ongoing efforts to optimize the data labeling workflow for its users, with the embeddings generated for all videos in the Catalog, including newly uploaded ones, available to both free and paid plan users.