Summary
Twelve Labs has released Pegasus 1.6, a vision-language model designed to turn first-person video into structured training data for robotics and physical AI. It analyzes actions, objects, tools and surrounding context, then divides a recorded task into labeled steps. Five core functions cover action segmentation and labeling, detailed descriptions, video-quality assessment, natural-language search and selection, and detection of material that may require privacy or compliance review, such as visible faces or documents. Its time-based metadata feature records events and context in a user-defined format. Twelve Labs says it improved consistent identification of hands, objects and tools, as well as processing speed and cost efficiency at scale. An unnamed US robotics-data company is reportedly using Pegasus to evaluate and annotate factory footage.
Hana’s perspective
Pegasus 1.6 shows how video understanding can create value without generating a single frame. Turning footage into timed, searchable structure is useful for robotics, but the same foundation matters in film and animation archives: teams could locate actions, flag continuity details, assess footage and build metadata at scale. The privacy review feature is especially important for first-person video, where faces and documents are easily captured unintentionally. Company claims about consistency, speed and cost still need independent evaluation on messy real-world footage.
Are you a writer with a great story? We'd love to hear from you: APPLY HERE