Vionlabs, a content intelligence company founded in Stockholm Sweden, wants to transform how audiences engage with content. They offer solutions to the global media and entertainment industry, assisting streaming services, studios, and broadcasters in understanding and enriching their video libraries. Vionlabs uses AI to extract deep metadata—including moods, emotions, environments, plot details, and textual nuances—from audio, video, and text content. This intelligence then powers improved content discovery, recommendations, and editorial workflows.

This decision was based on several factors, including maintaining high quality over a long context window; consistency in output that the model and its temperature parameter could achieve; and the ease of integration and cost-effectiveness of Vertex AI’s hosted API. Further, Vionlabs used Vertex AI to simplify launching and tracking new training jobs while also integrating seamlessly with its database of choice, BigQuery.

Vionlabs’ strategic choice allowed for rapid implementation, taking only a few weeks instead of the six to nine months typically required for training their own embedding models (the numerical representations or vectors that enable AI systems to perform semantic search, natural language processing, and cross-modal applications; and power recommendation systems).

Today, Vionlabs provides its clients with several new services powered by Llama 3.1 on Vertex AI, including:

• Fully automated, editorial smart lists: Vionlabs uses multimodal embedding for initial content clustering. Llama then refines these groups and generates list names, automating the previously manual and time-consuming process of curating content lists. This service helps clients organize massive libraries (e.g., 100,000 titles into 700 different lists) with compelling, auto-generated list names and descriptions. Clients benefit from more efficient content curation and the ability to instantly push these lists to their user interfaces for consumption.
• Frame-level, automated trailer creation: Leveraging its deep frame-level understanding of content, Vionlabs automates the creation of preview clips and short trailers. This service performs scene analysis, tracks key characters and story arcs, and identifies different moods. Llama contributes short synopses and tags for these automated clips, enabling clients to create highly customized assets, such as a stitched montage of all the high-octane action scenes from a cluster of action movies. In this way, clients can quickly create highly engaging promotional content.