VER 1.0 — The Visual Embedding Record

A Metadata-Resilient Standard for Image Embeddings

Status: Draft Standard, Version 1.0 · August 21, 2026 · Supersedes and incorporates the source thesis “A Metadata-Resilient Standard for Image Embeddings”

1. §0Critique of the Source ThesisThe thesis is architecturally correct.
2. §1ScopeThis document defines the Visual Embedding Record (VER): a serialization format, a canonical pixel normalization pipeline, an identity model, an embedding-space declaration mechanism, a metadata preservation and trust model, and conformance
3. §2Terminology and Requirement LevelsThe key words MUST, MUST NOT, SHOULD, SHOULD NOT, and MAY are to be interpreted as described in RFC 2119 and RFC 8174.
4. §3ArchitectureVER is organized as four layers.
5. §4Layer 0 — Canonical Pixel Normalization Pipeline (CPNP-1)CPNP-1 is asset-level and model-agnostic.
6. §5Identity ModelA Record MUST carry content_hash (SHA-256 of Asset bytes) and pixel_hash (§4 step 8).
7. §6Layer 1 — Canonical Visual Embeddings and Embedding Space DescriptorsEvery vector in a Record is bound to an Embedding Space by space_id.
8. §7Layer 2 — Metadata Preservation, Reconciliation, Trust, and PrivacyProducers at conformance level L1+ MUST preserve every metadata segment present at acquisition, byte-exact: { segment: exif|xmp|iptc-iim|icc|c2pa|other, bytes (base64), sha256 }.
9. §8Layer 3 — Contextual EmbeddingsContextual embeddings are optional, additive, and provenance-graded.
10. §9Provenance, Acquisition Chain, and C2PA BindingVER complements C2PA;
11. §10Record SerializationA Record is a UTF-8 JSON document conforming to the companion schema ver-record.schema.json (JSON Schema 2020-12).
12. §11Conformance LevelsM = mandatory, S = SHOULD, — = not required.
13. §12Security Considerations
14. §13Versioning and Extensibility
15. §14Relationship to Existing StandardsVER replaces none of these.