EBU Sovereign AI Framework Expands as Adobe Speeds Frame.io Video Search

Two different infrastructure problems are coming into focus for media teams using AI: who controls the systems that process sensitive material, and how quickly editors can find useful moments in large video libraries. At a Berlin creative industry event, the European Broadcasting Union (EBU) and NVIDIA described an expanded framework for European broadcasters, while Adobe detailed GPU-based improvements to video search in Frame.io.

The work addresses different parts of production. The EBU initiative is about deployment choices, data governance and security. Adobe’s changes target the processing pipeline behind search and media ingest. For broadcasters and production companies, the practical question is whether these systems can fit existing workflows while giving staff more control over where material runs and how they retrieve it.

EBU framework adds cloud and model options

The EBU and NVIDIA are expanding standards they introduced last year for European public service media. The new phase brings in sovereign cloud providers, open model developers and systems integrators. Its aim is to turn broad principles into reference architectures and requirements broadcasters can use when deploying agents.

The framework covers where workloads run, how data is governed, which models support services and how systems remain secure throughout their lifecycle. That scope matters as broadcasters consider agents for newsroom, production and audience services. Deployments in those settings may involve sensitive editorial material, local language needs and archives that organizations want to manage within their own operational boundaries.

The EBU represents more than 115 member organizations across over 55 countries, reaching over 1 billion people. Its chief technology officer, Annsofi Eriksson, said the shared standards are intended to help members strengthen resilience, retain control and deliver accountable services that sustain audience trust. The framework describes a common foundation; individual organizations will still need to make deployment decisions that fit their systems and requirements.

Two cloud providers are mapping their offerings to the initiative. Microsoft is contributing expertise from Microsoft Sovereign Cloud, including Azure Local and Foundry Local. The stated use cases include processing sensitive editorial material locally and keeping critical services running when external connectivity is unavailable. Nebius offers a platform for broadcasters to train, fine-tune and run their own models on European infrastructure.

Open models are another part of the approach. They give broadcasters the ability to inspect, evaluate and adapt models for local languages, archives and editorial policies. NVIDIA’s Nemotron models are cited as a foundation for domain-specific applications and public-service requirements. That flexibility can help teams tailor systems to their material, though deploying a model still requires decisions about evaluation, hosting and ongoing operations.

Delphine Ernotte, CEO of France Télévisions and EBU president, said sovereign cloud infrastructure and open models can support innovation while protecting data, editorial independence and cultural diversity. Qvest and other systems integrators are expected to provide media-specific consulting and integration, with the goal of making the reference architectures practical and interoperable in broadcast environments. Their role will be important in connecting these designs to the varied systems broadcasters already operate.

Frame.io targets faster retrieval from video libraries

Adobe’s Frame.io work focuses on the cost and speed of searching video at scale. Its GPU-accelerated pipeline uses NVIDIA NVDEC for decoding, TensorRT for optimized AI inference, and the CUDA platform for preprocessing and generating embeddings. Embeddings help represent video content in a form that can support searches based on meaning or visual content.

Adobe reports about five times faster embedding generation, up to four times higher throughput and up to five times lower latency for search processing. These figures describe improvements in the reported pipeline; the source doesn’t specify a benchmark dataset or provide the hardware configuration behind them. Adobe is also using NVIDIA NVENC and NVDEC for GPU-accelerated video encoding and decoding during media ingest.

A smaller proxy format is part of the search design. The new proxy runs at two frames per second and is about 80% smaller than the 360p version it replaces. Lower storage and transfer needs can make large libraries easier to process, while the proxy retains what Adobe describes as enough visual detail for search. The figures don’t establish how the change affects every production workflow or source format, but they indicate where Adobe is targeting efficiency.

UFA, a major German film and TV producer, provides a concrete example of the retrieval problem. The company delivers more than 3,500 hours of programming each year, including the daily drama “Gute Zeiten, schlechte Zeiten.” Editors creating a recap or flashback may need a particular reaction or an unresolved exchange recorded as far back as 15 years. Folder structures and file names often don’t describe those moments in enough detail to find them quickly.

Frame.io semantic search lets editors look for video moments by concept, action or setting. Hybrid search adds notes, transcripts and approval status, allowing an editor to combine visual queries with production information. For example, an editor can search for a concerned expression, narrow results by character name and restrict the results to approved footage before moving into review.

The workflow could reduce time spent locating candidate clips, especially in long-running productions with extensive archives. Search results still need to be checked by editors, and the source doesn’t report accuracy rates or explain how the system handles ambiguous queries. The reported speed improvements address processing capacity and response time; they don’t by themselves establish the quality of every match.

These announcements sit within a broader Adobe and NVIDIA collaboration that also includes Adobe Substance 3D, Firefly Foundry and enterprise 3D digital twin workflows. For media organizations, the clearest takeaway is that AI adoption depends on both the infrastructure behind a service and the details of how people use it: control over deployment for broadcasters, and faster discovery tools for teams working through large video collections.

Explore NVIDIA technologies for media and entertainment.

Read more about NVIDIA announcements in Europe.