Inferensys

Glossary

JUMBF

The JPEG Universal Metadata Box Format (JUMBF) is a universal container framework standardized in ISO 19566-5 that enables embedding any type of metadata, including C2PA provenance manifests, directly into JPEG and other file formats.
Data scientist building training data pipeline on laptop, data preprocessing visible, technical workspace.
Metadata Container Standard

What is JUMBF?

The JPEG Universal Metadata Box Format (JUMBF) is a universal container framework for embedding any type of metadata directly into JPEG and other file formats, serving as the foundational embedding mechanism for C2PA provenance manifests.

JUMBF (JPEG Universal Metadata Box Format) is a standardized, extensible container architecture defined in ISO/IEC 19566-5 that allows multiple types of metadata—from XML and JSON to binary data—to coexist within a single file. It provides a universal structure of nested boxes, each identified by a unique UUID, enabling applications to parse and ignore unknown metadata types without breaking file compatibility.

In the C2PA provenance ecosystem, JUMBF acts as the physical carrier for content credentials, embedding cryptographically signed manifest assertions directly into the asset's binary structure. This enables hard binding of provenance data, ensuring the tamper-evident metadata travels inseparably with the content across distribution channels without requiring external sidecar files.

UNIVERSAL METADATA CONTAINER

Key Features of JUMBF

The JPEG Universal Metadata Box Format (JUMBF) is the foundational container architecture that enables C2PA to embed any type of provenance data directly into media files. Its extensible, box-based structure is what makes content credentials interoperable across different file formats and applications.

01

Universal Box Structure

JUMBF defines a standardized box-based container that can encapsulate any type of metadata, regardless of its format or schema. Each box contains a type identifier, a payload, and optional label for human readability. This universal design means a single JUMBF container can simultaneously hold C2PA manifests, XML, JSON, CBOR, or proprietary binary data without conflicts. The structure is self-describing, allowing parsers to skip unknown box types while processing recognized ones.

02

Embedding in JPEG and Beyond

JUMBF content is embedded directly into JPEG files using the XMP and UUID boxes defined in the JPEG standard, ensuring backward compatibility with existing image parsers. Beyond JPEG, the format is designed for file format agnosticism, with defined embedding strategies for PNG, SVG, HEIF, WebP, and AVIF. This multi-format support is critical for C2PA's goal of universal provenance across all digital media types.

03

Superbox Nesting and Hierarchies

JUMBF supports superboxes—containers that hold other JUMBF boxes—enabling complex hierarchical metadata structures. This nesting capability allows C2PA to organize provenance data into logical groupings:

  • Assertion boxes for individual claims
  • Claim boxes that group related assertions
  • Manifest boxes that contain the full provenance record This hierarchical design mirrors the natural structure of provenance data, where a single asset may have multiple assertions from different actors.
04

CBOR and Binary Encoding

JUMBF payloads are typically encoded using CBOR (Concise Binary Object Representation), a binary serialization format defined in IETF RFC 8949. CBOR provides a compact, schema-less encoding that is significantly more efficient than JSON for embedded metadata. This efficiency is critical for hard binding scenarios where the provenance manifest must fit within the file's existing metadata allocation without bloating file size. CBOR also supports deterministic encoding, essential for cryptographic hashing consistency.

05

Extensibility Through Description Boxes

JUMBF includes a description box mechanism that allows new metadata schemas to be registered and identified without modifying the core specification. Each description box contains a UUID that uniquely identifies the type of metadata being carried, along with optional versioning information. This extensibility is what allows C2PA to evolve its assertion types—adding support for new kinds of provenance claims—without breaking existing validators or requiring updates to the JUMBF specification itself.

06

Labeling for Human Readability

Every JUMBF box can carry an optional UTF-8 label that provides a human-readable name for the contained metadata. This labeling system serves as a bridge between machine-parsable binary structures and human understanding, allowing debugging tools and forensic analysts to quickly identify the purpose of each box. Labels are not cryptographically significant but are invaluable for provenance inspection and validator development, making complex manifest structures navigable without decoding every binary payload.

JUMBF EXPLAINED

Frequently Asked Questions

Clear, technical answers to the most common questions about the JPEG Universal Metadata Box Format and its critical role in content provenance.

JUMBF (JPEG Universal Metadata Box Format) is a universal container framework for embedding any type of metadata directly into a JPEG or other compatible file format. It works by defining a standardized, extensible binary structure—a 'box'—that can hold multiple nested boxes of data, each identified by a unique label. This allows a single file to carry diverse metadata types, from XML and JSON to CBOR-encoded provenance manifests, without conflicting with existing file format specifications. JUMBF is the foundational embedding mechanism for the C2PA standard, enabling content credentials to be stored directly within the asset they describe, creating a self-contained, tamper-evident package.

Prasad Kumkar

About the author

Prasad Kumkar

CEO & MD, Inference Systems

Prasad Kumkar is the CEO & MD of Inference Systems and writes about AI systems architecture, LLM infrastructure, model serving, evaluation, and production deployment. Over 5+ years, he has worked across computer vision models, L5 autonomous vehicle systems, and LLM research, with a focus on taking complex AI ideas into real-world engineering systems.

His work and writing cover AI systems, large language models, AI agents, multimodal systems, autonomous systems, inference optimization, RAG, evaluation, and production AI engineering.