Image Collector
The Image Collector retains supported PNG, JPEG, and SVG images, including their bytes, identity, source revision, and digest. The same image can then appear in HTML and PDF after the original source is removed.
What it documents
Images illustrate Markdown pages and guides and published specifications. They can also supply a publication's product logo.
Include an illustration

Supported Markdown collection workflows collect the referenced image separately.
Repository Markdown can resolve a relative path such as
../assets/lifecycle.svg within the selected Git revision. Rendering restores
the retained bytes without fetching the image again. Supply meaningful
alternative text. SVG validation rejects active content and external references.
Use source-local images
A local website source tree can keep illustrations beside its canonical Markdown:

The website collector discovers SVG, PNG, and JPEG files beneath the documentation root and retains them as bounded binary assets. Relative Markdown image paths resolve from the Markdown file's directory; their canonical paths must stay inside that root. Missing files, paths escaping the root, query strings, and fragment identifiers fail explicitly. The same source file can supply website and PDF profiles, including selected section fragments, without fetching an absolute production URL. Rendering reads the validated bytes from the pinned KG.
A standalone local MarkdownFile uses its parent directory as the image root.
Embedders collecting standalone inputs can retain a corresponding image with
LocalImageFile::new(project, root, path).await?; the full website pipeline
schedules discovered images automatically. HTTPS requests keep their independent
collector and asset-factory behavior. Each encoded image may contain up to 8 MiB. A publication may retain up to
32 MiB of image bytes in total. Raster images must also fit within 8192
pixels on either edge and 16,777,216 pixels overall; decoding is capped at
64 MiB. These independent limits allow ordinary photographic illustrations
while rejecting oversized or compressed image bombs. Local and HTTPS collection
and rendering from an exported graph enforce the same limits. SVG-content
validation remains mandatory.
Configure a product logo
Publication definitions configure an explicit image job:
@prefix ga: <https://ekgf.org/ontology/grapharch#> .
@prefix dct: <http://purl.org/dc/terms/> .
@prefix ex: <urn:example:> .
ex:project ga:collector ex:logo-source .
ex:logo-source a ga:CollectorUse ; ga:collectorType ga:ImageCollector ;
ga:source "assets/product.jpg" ; ga:sourcePath "assets/product.jpg" ;
ga:asset ex:logo .
ex:logo dct:title "Example product" .
ex:specification ga:productLogo ex:logo .
This uses the configured publishing workflow of an
exact-revision build;
it is not a standalone --kind image CLI command or automatic family-build
discovery. Product logos are separate from authority logos and render at 100
CSS pixels in the specification header. Images referenced by configured
Markdown are collected from the same pinned commit without a separate job.
Interactive SVG navigation
Request interactive presentation when an SVG contains authored navigation links:

A website publishes this diagram in an isolated SVG object. Its source-owned
links are checked against the publication's routes and anchors, then projected
relative to the publication root. Links open the surrounding page. SVG styles
remain isolated from the documentation page. The publication theme controller
projects its selected color scheme into the SVG's authored
prefers-color-scheme media rules, including compound and nested queries.
The original SVG retained in the graph is unchanged.
The original SVG stays unchanged in the KG and remains available as an ordinary
image. A separate content-addressed SVG supplies website navigation. PDF uses
the same original diagram as a static image. Missing link targets, unsupported
SVG content, and non-SVG interactive images fail publication. Safe legacy SVG
<object> authoring receives the same interactive intent during collection.
Light and dark illustrations
Keep distinct authored variants together:
{.only-light}
{.only-dark}
Both assets remain in the graph. Website presentation selects the matching
color scheme; PDF chooses light and untagged images. Ordinary image titles are
preserved, including when presentation hints are combined. The explicit
grapharch:interactive-svg;only-light shorthand also combines interaction and
color intent.
Image references expose ga:imagePresentation and ga:imageColorScheme triples.
Normalized Markdown stores combined options and ordinary titles in typed
grapharch:image { ... } data so they survive export and restoration without
reopening authoring files.