Leahi, Sea and Ornament

The coast inside an ornamental interval

The pale beach curves toward a low, ridged headland beyond the sea. A botanical surround encloses the landscape while a large decorative disk rises behind the ridge. The scene alternates between the open interval of sky and water and the denser rhythm at its edges.

The reference photograph shows Diamond Head from Waikiki beach. It is an external coastal view with hotels and visitors, not a view into the crater. Removing visitors, reducing buildings and expanding ornament are authored choices. No measured landmark-recognition score or surveyed geographic reconstruction is claimed.

The guide retains conspicuous hotel blocks despite a request to reduce them, and the disk is much more ornamented than the intended quiet shape. Those departures are recorded rather than retrospectively described as fully satisfying the brief.

From a head-and-disk relation to a landscape hypothesis

The Foundation's Zodiac record describes a disk behind a depicted woman's head. Here the landmark takes the foreground role and a disk occupies the background. That is an explicitly authored cross-subject experiment, not evidence that Mucha historically used this relationship for Hawaiian landscapes.

Cleveland's Lorenzaccio provides a different relation: a figure and patterned field held within an ornamental border. The current picture adapts enclosure and botanical rhythm while omitting the actor, dragon, book, costume and lettering. A border inside the composition is not a physical frame.

The Library of Congress article by Melissa Lindberg describes botanical imagery, colouring and rounded/elongated shapes across selected poster designs. Those are qualified practice-level observations, not claims that every region of the1898 JOB poster or every oil painting uses one formula.

Design, variant and surviving sheet are different objects

Zodiac is identified as a1896 colour-lithograph design. The Foundation example is a text-free decorative-panel variant, while at least nine variants are reported. A calendar with lettering and a text-free panel should not silently become one identical reference object. No Zodiac image was admitted.

The eligible historical image is Cleveland's Lorenzaccio, accession1973.1018, dated1896-1900. Its broad margins, Les Maitres de l'Affiche PL114 inscription, Imprimerie Chaix credit and blind stamp identify the supplied reduced printed sheet. It is not presented as a full-size first theatre poster.

The1898 JOB design identified by the LOC caption is also distinct from the1896 design with the same product name. Related magazine and blog versions are one Melissa Lindberg source lineage. Cleveland's visual-description authorship is unestablished; that support is not labelled independent scholarly consensus.

What enters the proposed shared ontology

Ten assertions retain source spans, targets, quantifiers, qualifications and attribution across three identified works/designs and three delivery families. Shared concepts cover patterned surrounding fields and print-state context. New candidate distinctions include ornamental border enclosure, a disk behind an identified subject, and the combination of rounded and elongated forms.

School classification remains contextual. A named Art Nouveau work does not automatically inherit every motif. Likewise, attributed metallic colouring is an appearance description: it does not establish actual metal leaf, ink chemistry or angular reflectance.

The runtime ontology remains unchanged. Same-agent review is recorded, with independent semantic validation, held-out description scoring, minimality and explained variance still pending. Counting claims or shared identifiers is not an evaluation of how much artistic preference the ontology explains.

What the generator received

One built-in image-generation call received the saved concept brief, the eligible landmark thumbnail and Cleveland's eligible reduced Lorenzaccio sheet. Their distinct reference roles are preserved in the prompt. The source composition is not copied, and no historical image of Zodiac was supplied.

The landmark image carries Qccontrib's individual CC0 dedication. The original download failed; the provider's1280pixel thumbnail was acquired and decoded. Its lower resolution and existing photographic rendering constrain fine visual inference.

Cleveland's exact object has CC0 image eligibility. Its embedded colour profile was converted to sRGB without cropping or resizing, with the original file and conversion receipt retained. Display normalization does not recover original historical inks or prove physical print reproduction.

The artist name, concept text, reference images and generator prior knowledge all contribute to this setup. Their separate causal effects remain unmeasured; an artist-name or concept ablation has not been run.

The recorded painting beneath the panel

The native canvas contains 15840 stored marks. Four base passes deposited 3989 strokes. A further 18000 refinement proposals added 11851: 11020 gradient-colour and 831 solid-colour marks. The other 6149 proposals deposited no paint.

The generic engine fits local direction and colour alternatives to a resized guide and accepts marks when local squared RGB error falls. Refinement reduced mean squared error by 59.5 percent. This is image reconstruction, not a learned lithographic process or an independently validated Mucha hand policy.

Restricted standalone replay reproduces the native painting exactly. The full film records every accepted mark, with a shorter accelerated preview. Marks are deposited atomically; no physical hand or continuous brush-tip trajectory is simulated.

Where the transfer holds at coarse scale

The border still encloses the coast and the disk remains behind the headland. Those large spatial relationships are qualitatively retained in this same-agent comparison. Fine stems, petal outlines and small interior details are softened or merged.

The native renderer introduces dense curled and faceted microtexture, including thick-looking highlights. The intended restrained print-like handling is not established. Low reconstruction error cannot establish lithographic ink deposition, historical paper interaction or precise contour fidelity.

The image-hash-bound control review labels only the coarse relations as qualitatively retained and leaves fine pattern/contour handling unvalidated. There is no independent historical-style review. Five CNN/ViT observations include the historical reference, but their mixed image features do not isolate pure style or validate a print process.

The next test should separate enclosure from surface

Hold the coastal composition, palette and resolution fixed while comparing the generic renderer with one that explicitly preserves thin outlines and suppresses unnecessary interior texture. Score border continuity and contour fidelity separately from landmark recognition and preference.

A second comparison could keep the same landscape and border while removing or changing the disk. The disk-to-subject relation should be tested independently of generic circular prominence. Proposed tests remain unexecuted.

A fresh AIC Woman with Daisy record is reserved unread for later identity/medium verification and evaluation. Reservation does not imply exclusion from model pretraining or earlier research. This artwork and its scoped evidence are a development result, not a validated all-career artist genome.