MachineTradecraft
Global site search

Search guides, labs, glossary, and research

Type two or more characters to search.

Start with a channel, artifact, or defense term

Examples include zero-width, metadata, tokenizer, or prompt injection.
    Media & Provenance · Cross-modal conflict

    Multimodal Image Differential

    The lab distinguishes metadata extraction from OCR and uses only bundled deterministic OCR text for prepared fixtures.

    Browse all labs Open universal Machine View
    Quick answer

    What does the Image Differential show?

    The lab distinguishes metadata extraction from OCR and uses only bundled deterministic OCR text for prepared fixtures.

    Human visibility
    Pixels, alpha, descriptions, and metadata
    Machine receiver
    Local image parser and prepared OCR fixture
    Robustness
    Image and preprocessing dependent

    Research boundary: this page uses bounded, inert data and fixed safe examples. It never executes decoded content, requests secrets, calls third-party services, or performs actions against external systems.

    FIRST RUN / THREE STEPS

    Start with the prepared, bounded workflow.

    Nothing runs automatically
    1. Choose one supported file

      The file stays in PHP’s temporary upload for this request and is not retained by the application.

    2. Run Inspect image representations

      Inspect the selected file as inert bytes and containers. No embedded content is executed.

    3. Scan before expanding

      Read the summary first, then scan findings and expand only the machine views you need.

    INPUT / CONTROL PLANE

    Prepare the input and choose one action.

    Ready

    The recommended first run is separated from alternate analyses. Inputs and selected files stay on this host.

    Current input state No file selected yet

    Choose one supported local file, then run the recommended inspection. Browsers do not repopulate file selections after a reload.

    Maximum 4 MiB. No image, OCR, or model request leaves this host.

    Maximum file size: 4 MiB. Accepted file types: PNG, JPG, JPEG.

    ACTION HIERARCHY

    Run the recommended first pass.

    Alternate actions remain available below, but the first pass is the clearest place to start.

    Inputs remain on this host. Text operations are size-limited; uploaded files are processed from PHP’s temporary upload and are not retained by the application.

    OUTPUT / MACHINE VIEWS

    Scan the result from summary to evidence.

    Run Inspect image representations to create the first result.

    Choose a supported file first. The output will lead with a summary and visible qualifications before the expandable machine views.

    SummaryFindingsMachine views
    Interpretation framework

    The same artifact can produce several valid observations.

    01

    Human view

    What a person naturally reads, sees, or hears.

    02

    Structural view

    What a parser, DOM, container reader, or metadata extractor exposes.

    03

    Decoder view

    What becomes meaningful only with a rule, key, tokenizer, model, or tool.

    04

    Defensive view

    What normalization, rendering, OCR, canonicalization, or policy changes.

    Evidence and decision boundary

    Use the result as bounded evidence, not as a universal verdict.

    The lab distinguishes metadata extraction from OCR and uses only bundled deterministic OCR text for prepared fixtures.

    LOCAL MODEBounded image-container inspection with prepared OCR boundary
    REVIEW DATE2026-08-26
    SOURCE BODYPreserved separately from implementation claims
    Computed locally

    Deterministic output produced by this bounded runtime.

    • Exact digest, supported image dimensions, PNG/JPEG metadata, selected chunk or marker information, and basic transparency indicators
    • Whether a PNG declares an alpha-capable color type or contains a tRNS token
    • No uploaded pixel content is sent to an OCR engine, vision model, QR resolver, or provenance service
    Bounded approximation

    Useful subset or model that does not establish full conformance.

    • The prepared OCR explanation is not OCR of the uploaded image
    • A string-level tRNS check and color-type check do not characterize all compositing or display behavior
    • No visual description, adversarial-example detector, watermark detector, or model inference is produced
    Escalate for

    Claims that require an exact parser, trust system, model, or human review.

    • OCR requires a pinned engine, language data, preprocessing, layout mode, confidence output, and representative ground truth
    • VLM behavior requires the exact model and preprocessing pipeline in a side-effect-free test
    • QR, watermark, and C2PA checks require dedicated bounded decoders or verifiers
    Interpretation rule

    Record the receiver and transformation.

    Machine-decodable is receiver-relative. Record the parser, preprocessing, codebook, tokenizer, key, model, and transformation path before generalizing from one result.

    Limitations

    What this page does not prove

    The deployed runtime does not perform live OCR or vision-language inference. Prepared OCR examples explain representation conflicts without claiming model behavior.

    Deterministic review material

    Download the exact benign fixtures used for the evidence boundary.

    These local files are supplied for repeatable inspection. The application does not fetch them automatically, execute their content, or treat a fixture result as external verification.

    Multimodal metadata PNG

    Benign PNG metadata fixture for comparing container metadata and transparency indicators without OCR or model inference.

    Type
    PNG fixture
    Bytes
    9,527
    SHA-256
    f0873af15cf4eb188bdb…
    Download fixture

    Download the complete evidence-review fixture pack · SHA-256 sidecar included beside the archive

    Continue the investigation

    Related labs and source reports

    Provenance Inspector Structural parser and verifier Multimodal Machine Perception: Pixels, OCR, Alternative Text, Metadata, and Visual Instruction Security Research report Accessibility and Image Metadata in Multimodal AI Systems Research report Content Provenance, Content Credentials, Watermarks, and Authenticity Signals Research report
    Continue exploring

    Move from the current explanation to a related guide, evidence source, or controlled workbench.

    Laboratory catalogChoose a focused local experimentBrowse 38 bounded laboratories spanning representation, files, messages, models, provenance, agents, supply chains, and defense.Open this path Field guideDefine Machine Tradecraft preciselyUse the formal model, classification test, taxonomy, and limits before generalizing from one artifact.Open this path Defensive architectureBuild a safer model-ingestion boundaryPreserve evidence, compare representations, canonicalize allowlisted fields, retain provenance, and minimize privilege.Open this path