Para Bellum

The work and its context

Para Bellum makes an offer that it cannot quite fulfil: a picture that looks ready to be read. Large letters cross painted-looking rectangles, sit inside them or disappear at their edges. Familiar typographic forms promise a message, but the strings they spell resist that promise. The eye alternates between trying to recover a word and accepting its letters as coloured shapes. In the five works examined here, that change of attention is as consequential as any change of palette. ^{1 2 3 4 5}

Matty Mariansky's project was released through Art Blocks Curated on 7 February 2022, with an edition of 1,000. His statement assigns emotion to colour and reasoning to language, bringing them into conflict within one image. It also describes a language engine informed by books about rebellion and anarchy. The title alludes to a Latin maxim connecting the desire for peace with preparation for war. These are the artist's declared premises; the picture need not illustrate a particular battle to make conflict visible. ⁶

The difficulty is that lettering here never remains purely rational. Cropping, transparency and unfamiliar combinations turn it into an uncertain visual material. Conversely, the fields of colour have divisions, proportions and a deliberate order of assembly. My reading is that the pictures complicate their announced opposition: both systems are calculated, and both can frustrate recognition. AB5D's #659, with its diagonal ochre letters crossing red blocks, makes that instability especially clear. The Matty Mariansky dossier provides the wider context for his practice. ^{7 5}

Conception and development

In a contemporary interview with Jeff Davis, Mariansky connected the language experiments to his interest in antique amulets and the imagined efficacy of particular letter combinations. He tested generators small enough for on-chain use while seeking strings that approached words without settling into them. He identified Rothko as an influence on the colour fields, then described the emerging resemblance to Swiss posters as a reason to choose a portrait canvas. These are documented references, rather than influences inferred solely from appearances. ⁸

The interview also describes an earlier practice of variable, animated club flyers made with Flash. That history helps explain why typography and algorithmic variation meet so naturally here. Mariansky encouraged viewers to consider combinations of outputs and personally suggestive strings rather than concentrating on statistical rarity. The interview establishes this development account, although the linked technical page on his own website was unavailable during research. ⁸

What survives in the deployed program is more specific than a general ambition to generate language. It stores character relationships and the outlines needed to draw the resulting letters. It also specifies several distinct ways of placing text, from large isolated forms to dense rows. The project therefore authors both an alphabet's material appearance and the conditions under which that alphabet becomes difficult to read. A new string changes the image because different letters occupy different widths and enclose different spaces, even before any possible verbal association is considered. ⁷

How the work was constructed

The complete indexed program uses p5.js 1.0.0. Its random system creates two streams from the two halves of the token hash, advances them through a warm-up sequence, then alternates between them. Subsequent choices determine colour, composition and lettering. This gives the image a reproducible input without requiring the artist to arrange each token separately. The inspection concerns the retrieved indexed script; a pinned retrieval directly from the contract remains a separate archival task. ^{6 7}

The language mechanism is a table of probabilities for the next character, conditioned on the preceding character. Starting from a space, it chooses letters and capitalises the first one. An early space can be skipped while generation continues; a space after sufficient letters ends the string, and a length limit also stops it. This explains why some sequences resemble fragments of English without following ordinary word boundaries. The deployed mechanism does not consult an external language service or contain a neural model. Its compact table is enough to produce the effect. The statement's account of the books used to prepare it does not supply the original corpus for independent reconstruction. ^{7 6}

The font is equally concrete. Instead of asking the browser to download a typeface, the script contains outlines and advance widths for upper- and lower-case glyphs adapted from Inter. A parser turns their path instructions into vertices and quadratic curves. Letter spacing, scaling and line wrapping are then handled within the drawing procedure. The type is consequently part of the image-making program, not a separate layer of ordinary webpage text. Its contours can remain crisp even when the forms behind it imitate uneven pigment. ⁷

A logical canvas of 1,000 by 1,200 units establishes the portrait format and is scaled to fit the viewing window. Some compositions use nearly the full field; others reserve a square above or below a strip of text. The colour structures have two principal construction routes. One repeatedly subdivides a field into rectangles. The other grows outward from an initial rectangle, with subdivision also available, and recentres the resulting group when necessary. These routes can produce very different distributions of occupied and open space without changing the basic vocabulary. ⁷

The textured treatment adds more than a rough outline to each rectangle. It builds irregular polygons, repeated narrow strokes, small variations of colour and occasional marks that read as circular sweeps. Grain modifies the pixel field, while blur contributes to the softer ground used in some compositions. A plainer branch can instead draw flatter blocks. The program contains 37 palette arrays, but that inventory is not a measured account of their prevalence in the minted edition. The five examples below all have textured treatments and cannot establish the distribution of the alternative. ^{7 1 2 3 4 5}

Text placement introduces another range of possibilities. The main strategies include a large word, rows of different words, a single positioned word, a paragraph of capitals, repeated text with changing brightness, and an oversized overlapping string. Certain configurations rotate the lettering. Others change its colour letter by letter, blend it with the existing field or erase through a layer to expose a saved background. Wrapping and clipping mean that an entire generated string need not remain visible. This is why a trait identifying a text strategy cannot replace looking at the resulting picture. ⁷

The final ordering is particularly important. Colour fields precede the main lettering, but further rectangles can cover parts of the text. There is no permanent division between a passive background and an authoritative foreground. A letter may cut across one block yet disappear beneath another. The eye's difficulty in deciding what belongs above what is grounded in these actual changes of drawing order, rather than in a simulation of sheets of paper, paint thickness or physical transparency. ⁷

Rendering proceeds through a finite sequence and ends with noLoop. In AB5D's #659, an initial red ground was observed becoming a completed arrangement of blocks and diagonal letters; later observation found the finished image unchanged. This is a progressive drawing that reaches a still, rather than a continuously animated work. No custom pointer, keyboard or audio implementation was found. The script also contains browser-specific blur handling and a mobile pixel-density adjustment, so exact pixel equality across display environments should not be assumed. ^{7 5}

The mint

The first indexed token was minted on 1 February 2022 at 17:38:30, before the public release date given on the collection page. Activation is recorded on 3 February. The scheduled opening was 7 February at 18:00:00, the second indexed mint at 18:16:55, and completion at 19:08:25. Full timestamps and offsets remain in the source record. The earlier first token and the later release describe different events; substituting one for the other would obscure the sequence. ^{6 1 2}

The current record contains 1,000 invocations and a maximum of 1,000, consistent with the collection page's stated edition. It also contains a price setting of 0.1 ETH. That setting does not establish the original public price, any intervening changes or the amount paid for each mint. Likewise, the interval between the second token and the completion field does not independently identify a public sellout duration. Allocations and transaction classifications have not been reconstructed here, so the essay retains the chronology without turning those fields into a more precise commercial narrative. ⁶

Copyright and licensing

The project record reports a combined label: CC BY-NC 4.0 for the artwork and an Open Font License notice for Inter. Its terms-link field is empty. The named Creative Commons deed permits sharing and adaptation subject to attribution and the noncommercial condition, with changes identified where applicable. This describes the observed artwork label; the history of any earlier project terms has not been established. AB5D's dedication of this essay to CC0 applies to the original prose, not to Mariansky's images. ^{6 9}

The components have separate notices. The script credits modified SVG code to Robert Cook and Peter Collingridge under MIT, and its adapted font to Rasmus Andersson and contributors under the Open Font License. Inter's upstream licence specifies SIL OFL 1.1; its requirements for font software do not impose that licence on every document made with the font. The p5.js 1.0.0 repository supplies LGPL 2.1 for the library. None of these component notices establishes a blanket licence for the complete artist program. The exact font revision and upstream SVG version remain unresolved, and this publication retains a digest and analysis rather than redistributing the artist's script. ^{7 10 11}

Close reading of five outputs

The comparison uses #0, #1, #10 and #20 as a predetermined early group, plus AB5D's #659. The originals were opened in Chrome and inspected as rendered artworks. All belong to project 261 on the contract identified in the source record; #659's indexed owner matches an account in AB5D's complete linked-wallet response. This is a small comparison chosen for sustained looking, not a representative survey of all 1,000 outputs. ^{1 2 3 4 5}

In #0, an ochre field carries red and blue blocks, circular brush-like marks and several rows of large mixed-case lettering. Some letters stand out sharply in blue; others almost merge with the surrounding gold. The left edge clips the beginning of several rows, so the viewer encounters text already in progress. Its broad horizontal movement contests the vertical edges of the rectangles. Where similarly coloured letter and block meet, their boundary becomes difficult to retain: a typographic stroke can momentarily seem to belong to the painted structure behind it. ¹

The circular marks complicate that reading further. Their visibly built-up strokes suggest a different tempo from the smooth glyph contours, even though both come from the same program. Looking at the picture as a whole, the lettering distributes attention across the page; looking closely, a single blue letter or red patch can interrupt the attempted sentence. The result is not simply illegible writing placed over an attractive background. Each layer repeatedly changes the visibility of the other, and the failed reading becomes part of how the composition is experienced. ^{1 7}

In #1, pale cyan capitals fill a much denser arrangement of lines. Red blocks occupy much of the right side, a dark brown region weighs down the lower left, and a blue circular mark sits to the right of centre. The letters keep a consistent enough scale to suggest a paragraph, but the line endings and unfamiliar strings prevent an ordinary passage from emerging. Small blocks intrude over individual letters. Those local interruptions are forceful because the typographic rhythm initially looks so regular. ²

The cyan has a different relationship to each ground. Against red it feels bright and declarative; against pale blue it becomes comparatively slight. A single text layer therefore does not maintain a single visual volume across the sheet. Near the lower edge, another blue rectangle covers part of the writing, making colour behave like a correction or an obstruction. That analogy is my reading, not evidence that the artist intended a censored document. What can be observed directly is the alternation between crisp letter edges and areas where colour interrupts their continuity. ²

In #10, the balance shifts strongly toward the fields. Red and ochre rectangles occupy nearly the entire portrait, with dark margins and a pale vertical strip descending through the lower centre. Circular sweeps disturb the predominantly upright arrangement. Only a fragment of dark, rotated lettering enters prominently from the bottom. The index identifies a fading-lines configuration, but the visible result is not a page neatly displaying repeated rows. Position, scale and clipping have made that strategy largely peripheral. The discrepancy is an instructive reason to treat metadata as a description of selected parameters rather than a substitute for the image. ^{3 7}

In #20, warm capitals swell across the upper half of a red and orange field. Their size turns counters and vertical strokes into major compositional shapes. Broad red areas lower down remain comparatively unlettered, while a small yellow dot beneath the central type becomes conspicuous through isolation. Here the text does not create the dense, even reading rhythm of #1. It operates more like a partial sign encountered too close to see whole, with the surrounding fields setting the scale of that encounter. ⁴

AB5D's #659 takes the opposition into a diagonal register. Ochre and magenta capitals cross a patchwork of red and burgundy, leaving fragments at all four edges. A large rounded letter near the middle is legible as a shape before its place in any string can be found. The blocks retain vertical and horizontal boundaries, while the text cuts across them at an angle. That disagreement gives the page its pressure: two systems organise the same surface in incompatible directions. ⁵

Its colours also keep the layers from separating cleanly. Some ochre forms look solid against a dark rectangle, then become less distinct as they cross a brighter patch. Magenta strokes near the edges compete less successfully with the red ground, appearing to recede. A pale block in the upper-right area and thin marks elsewhere further interrupt the lettering. The picture offers enough familiar alphabetic structure to invite decoding, but its strongest coherence arrives through the crossing directions, contrasts and intervals that remain when decoding fails. ⁵

Significance and limitations

Para Bellum gives generative typography a substantive task. It does not merely change the wording of a poster while preserving a conventional hierarchy. It varies how much authority writing can hold at all: dense and insistent in #1, almost absent in #10, oversized in #20, diagonally entangled with colour in #659. The common program sustains those differences through the interaction of character generation, glyph geometry and the ordering of marks. ^{7 2 3 4 5}

The series is most persuasive when the viewer stops demanding that one layer explain the other. A colour field does not translate an invented word, and a word does not supply a reliable caption for the field. Instead, attempted reading alters the pace of looking. That specific relationship between verbal expectation and visual attention is the work's contribution to this part of the AB[500], and it remains available in individual pictures without a ranking of their traits. ^{1 2 5}

Unresolved research questions:

• Recover the original language-training corpus and the process used to derive the character table.
• Locate an accessible copy of the artist's detailed project-development page and compare it with the deployed program.
• Establish historical opening prices, any price changes and the division between public and other mints.
• Identify the precise Inter and SVG component versions and any separate licence for the complete artist code.
• Compare the indexed script with a pinned chain retrieval, test browser-specific rendering differences and extend the visual sample beyond these five works.