Versos AI Adds Plain-Language Curation for Licensed Video Training Data

Versos AI has launched a workflow that lets AI teams describe needed video training data in plain language, then turns that request into structured search and curation criteria. The system is designed to find, assess and assemble footage prepared for controlled AI licensing into a dataset.

One footage brief, many constraints

Video-data requests can combine what must appear on screen with technical and legal conditions. Versos says its workflow is designed to handle subjects, actions and environments alongside duration, resolution, format, language, production quality and permitted use. The company says meeting a detailed brief can otherwise require multiple searches, manual inspection and dataset preparation.

The workflow’s stated inputs

• Content requirements, including subjects, actions and environments.
• Technical requirements, including duration, resolution, format and language.
• Production-quality and permitted-use requirements.

Rights information is part of the search

The distinguishing feature is not simply prompt-based video search. Versos says the workflow searches footage prepared for controlled AI licensing and maintains its associated ownership, licensing and provenance information. That places documented permissions alongside relevance and technical fit when the system selects material for a training dataset.

The initial technology stack and next demonstration

Versos says NVIDIA CUDA Toolkit supports inference in its Video Library Intelligence Platform, NVIDIA Nemotron Ultra runs the agents, and LangChain coordinates the workflow. The company says its architecture is model-agnostic, allowing other open-weight or frontier models when customer performance, cost or deployment requirements differ.

Versos plans to demonstrate the capability at IBC2026 in Amsterdam from September 11 to 14. The announcement describes the intended workflow, but the company has not presented independent evidence here about how well it performs across customer datasets or licensing requirements.