Autology
steganon · Project 209 · Curated
Edition size: 1,024 · First mint: 11 November 2021

The work and its context
Autology makes two different kinds of structure available in the same image. On its surface, nested polygons gather into fans, ribbons and cavities. Within its pixel values, the image carries text that encodes its own generating program. The visible and encoded structures invite different forms of attention: one can be followed with the eye, while the other requires a process of extraction.12
The series contains 1,024 works and belongs to Art Blocks Curated. Its project number is 209. The official collection description presents the embedded source as integral to the work, rather than as an accompanying document. The steganon dossier provides context for the artist's wider practice; this essay concentrates on the particular relationship between image, program and reproduction in Autology.3
Five original renders were examined: editions 0, 1, 10, 20 and the AB5D-held #309. They reveal substantial changes in balance and apparent material. Some suggest ornament organised around a central axis; others spread restless fragments across the square. These are visual interpretations of the inspected examples, not claims that the program depicts a specific object or that this small group represents the edition statistically.24567
Conception and development
Steganon's own development account connects Autology to an earlier interest in computer security. A university exercise involved hiding a team's source code in an image. Years later, discovering Art Blocks brought that experience into contact with generative art. The artist describes more than six months of work on Autology, with the encoding mechanism established before the final visual language.8
An initial project called Recursion pursued recursive procedures and images resembling repeated reflections. That direction did not satisfy the artist as a long-form system. Studying Manolo Gamboa Naon's work led through Delaunay triangulation to Voronoi diagrams. Structured point arrangements proved more promising than unrestricted random placement. A mistake that removed the outer polygon opened spaces between cells and helped establish the eventual aesthetic.8
This account locates the project at the meeting of two investigations: an image that carries its own generating instructions, and a geometric system whose gaps and internal layers invite visual interpretation. The two need not be confused. The hidden program is recovered numerically; the visible forms remain available to viewers who never attempt extraction.18
How the work was constructed
Autology joins a geometric drawing procedure to a second procedure that stores text in the resulting pixels. The two operations are distinct. The first produces the visible composition; the second changes the image data so that it also carries a program. The source is therefore part of the image's numerical representation without appearing as ordinary printed text on its surface.1
Four unsigned integers parsed from the token hash initialise a bitwise pseudorandom sequence. The same sequence supplies decisions about layout, point placement, palette and later pixel treatment. A helper averaging three draws favours middle values for some choices. The composition is designed in a thousand-unit square and scaled to the smaller browser dimension, so geometric coordinates and output pixels are related but not identical.1
The drawing starts by placing sites. Several procedures arrange points along circles, spirals, cubic curves, rotated segments and grids. Some introduce further structures inside a chosen region, and some reflect a region's sites. A layout can use a regular subdivision or group neighbouring cells into larger rectangular areas. These operations matter because the final polygon boundaries are not drawn directly from a catalogue of shapes: they emerge from the relationships among these prepared points.1
An embedded Voronoi implementation divides the square according to those sites and clips the resulting cells to its boundary. The library carries a Raymond Hill copyright notice and an MIT licence reference. The artist's contribution includes the site-generating procedures and the subsequent treatment of the cells. A Voronoi diagram alone would not explain the particular rhythms, symmetry or colouring selected by the surrounding program.1
Colour is assigned by traversing neighbouring cells, selecting from one of eleven stored palettes. The drawing routine then renders nested versions of each polygon around its site. A scale factor is repeatedly squared, pulling successive layers inward, while lightness increases between layers. Different scale rules produce different relationships between a cell's outer boundary and its nested interior. An optional margin reduces the field, and boundary checks can suppress shapes that would reach beyond that permitted area. The visible result consequently depends on both adjacency and the internal treatment of individual cells.1
After drawing, the program reads the canvas pixels. It constructs a text payload containing the token hash and the self-referential source wrapper, prefixes its binary character data with a thirty-two-bit length value, and writes successive bits into the least significant bits of red, green and blue channels. The loops traverse pixels column by column. Other small, seeded changes to channel values add a noisy surface, and alpha is set to full opacity. This is a precise encoding operation, not a metaphor for an image that merely resembles software.1
The least significant bit has a small numerical influence on a colour channel. Replacing it can carry binary data while making only a slight local colour change. Recovering the stored text nevertheless depends on retaining those exact pixel values. Rescaling, colour transformations or lossy compression can change the bits on which recovery relies. A screenshot or a recompressed preview is therefore not automatically interchangeable with the original exported PNG as an encoded artefact.1
The script performs its drawing during initialisation and contains no continuing animation loop. A key-up handler for S saves the canvas as a PNG named after the hash. No pointer-based drawing control or audio routine was identified. A separate test serialised the original #309 canvas as a PNG, decoded that PNG into an image, and extracted the column-first RGB bits using the inspected decoder's scheme. At 720 by 720 pixels, it recovered all 17,367 characters. The recovered text's SHA-256 matched the indexed program prefixed by this token's hash assignment. This verifies preservation of the payload through that PNG round trip; it does not test every resolution or execute the extracted program anew.1
The mint
The index records the first token on 11 November 2021 at 04:59:08, activation on 18 November at 04:59:19, and a scheduled start on 22 November at 18:00:00. The second token is recorded at 18:22:15 on 22 November, with completion at 19:02:08. These fields describe separate events. The early first token does not establish that the public sale opened eleven days before the scheduled start.91
The scheduled-start-to-completion interval is one hour, two minutes and eight seconds. The second-mint-to-completion interval is thirty-nine minutes and fifty-three seconds. Neither calculation independently reconstructs a continuous public sale. Transaction classifications and purchase receipts would be needed to distinguish reserved or administrative activity from public purchases, and to establish what each recipient paid.91
The fresh index records 1,024 invocations against a current maximum of 1,024, a price setting of 0.5 ETH, and an empty sales-notes field. The price is a configuration observed during this research, not independently verified evidence of the opening price or of every mint's consideration. The edition is complete in the current record; an account of the original sale requires additional launch-era evidence.9
Copyright and licensing
The project's current artwork label is CC BY-NC-ND 4.0. The named Creative Commons deed permits sharing under attribution and noncommercial conditions; adapted material may not be distributed under that licence. The distinction is more precise than saying that the licence prohibits every act of making an adaptation. The index supplies no separate terms link, and an earlier history of the artwork's terms has not been reconstructed here.910
The embedded Voronoi implementation carries Raymond Hill's copyright and an MIT licence notice. That permission belongs to the library and does not by itself license steganon's surrounding program or the resulting images. No separate artist-code reuse grant was identified in the inspected source. Recoverability of code from pixels is a technical property, not evidence of an unrestricted reuse licence. AB5D's CC0 dedication applies to this essay's original prose.1
Close reading of examples

The comparison uses editions 0, 1, 10 and 20 as an early-edition group, alongside #309 from the AB5D collection. Each was viewed in its official generator. Their token identifiers combine project 209 with the respective invocation; full identifiers and links are retained in the example records. This selection supports close comparison of individual compositions, not an estimate of trait frequencies.24567

In #0, a mustard ground separates cream, grey and near-black pieces. The upper half is dominated by tall slats that narrow toward a small central knot. Some lean inward like ribs; others remain almost vertical. Their changing widths make the centre feel compressed, as though a broad structure were being drawn through a narrow opening. Two dense radial formations occupy the lower half. Their tiny interior cells expand into broad outer wedges, so the eye encounters a continuous change of scale within each cluster.2
The upper and lower systems do not settle into one object. The long slats suggest depth and convergence, while the paired fans can be read as flat arrangements turning around empty centres. The mustard channels between them maintain their separation. Colour also unsettles the bilateral geometry: dark pieces on one side can answer lighter ones on the other, preventing symmetry from becoming simple duplication. The work feels ordered without giving every corresponding part equal visual weight.2

Edition #1 makes that distinction between geometric and chromatic organisation even clearer. On a green ground, blue, maroon, orange, yellow and nearly black cells form repeated paired cavities. Near the top, rounded interior pieces seem suspended inside a larger broken ring. Across the middle, two radial openings establish another pair. Lower down, yellow and orange dominate a broad band around smaller circular centres. The eye can follow a vertical axis through the image, yet the changing colour distribution makes each level feel different.4
Narrow striped ribbons curve around those larger openings. Their close spacing creates a different pace of looking from the broad wedges: one reads as a vibrating band, the other as a set of separate pieces. The lower yellow region appears more expansive than the darker upper regions, although both belong to the same square. This is a useful counterpart to #0 because it shows how a related vocabulary can produce a comparatively buoyant, patterned surface instead of a compressed, architectural one.4

Edition #10 replaces that paired stability with an orange field crossed by grey, white and black fragments. Large pale fans project diagonally, while small rings and jagged clusters gather toward the upper part of the square. A broad open region near the lower middle lets the orange ground become an active shape. The white strips at lower right are especially forceful: their widening intervals make them seem to unfold into the empty field. The composition's energy comes from unequal concentrations and directional changes rather than from a single balancing axis.5

Edition #20 is quieter in colour but more insistent in its repeated structure. Black and grey forms sit on a pale ground with a clear outer margin. Two large circular cavities near the top answer one another; broad fans open toward either side near the middle. Horizontal ribbons cross between these events, linking regions that might otherwise read as separate emblems. A denser lower central cluster gathers the eye again. Without saturated colour to separate motifs, differences of spacing and scale become especially conspicuous.6
AB5D's #309 presents red, muted teal and lavender-grey cells on a pale blue ground. A tight spiral near the upper left contrasts with a larger radial opening toward the right. Meandering narrow bands connect and interrupt the broader polygon groups, and a small rosette near the lower middle supplies another centre of attention. The bright red segments repeatedly pull forward against the cooler colours. It is an airy arrangement despite its many components, because the blue channels remain wide enough to distinguish one movement from the next.7
Across these examples, gaps do more than reveal a background. They determine whether a group reads as one fan, a chain of individual tiles or an opening surrounded by fragments. The nested interiors can resemble bevelled material, but that impression remains an interpretation of flat colour layers. The code's geometric construction explains how the pieces are related; it does not exhaust the shifting associations that arise while looking.127
Significance and limitations
Autology's particular achievement is to give a picture two forms of continuity. The visible composition persists through relationships among its cells, while the encoded payload proposes a way for the image to retain instructions for its own generation. These are connected by the same program, but they are not the same kind of evidence. Seeing a convincing pattern does not establish that its pixels still contain a recoverable payload.1
The distinction makes the ordinary act of saving an image unusually consequential. A visually similar reproduction could preserve the composition while changing the colour-channel values used by the encoding. Conversely, an intact payload is not a substitute for attention to the individual image. The comparison between #0's compressed slats, #10's open orange field and #309's cool channels demonstrates why the artwork deserves to be examined as a composition as well as a technical proposition.1257
Unresolved research questions:
- Extend the verified 720-pixel PNG extraction test to other dimensions and independently test regeneration from the extracted program.
- Compare decoded payloads from original exports with those in platform preview assets.
- Recover launch-era price and edition-limit records and classify the first mint.
- Locate any separate artist-code licence and earlier artwork terms.
- Compare the indexed script against a pinned on-chain retrieval.
- Extend visual comparison beyond these five editions.
References
-
Art Blocks. tokens and script 209. Accessed 2026-09-18. ↩
-
Art Blocks. Autology #0. Accessed 2026-09-18. ↩
-
Art Blocks. Autology collection. Accessed 2026-09-18. ↩
-
steganon. Autology #1. Accessed 2026-09-18. ↩
-
steganon. Autology #10. Accessed 2026-09-18. ↩
-
steganon. Autology #20. Accessed 2026-09-18. ↩
-
Art Blocks. Autology #309. Accessed 2026-09-18. ↩
-
steganon. Autology: the backstory and decoder. Accessed 2026-09-18. ↩
-
Art Blocks. projects_metadata 209. Accessed 2026-09-18. ↩
-
Creative Commons. CC BY-NC-ND 4.0 deed. Accessed 2026-09-18. ↩
Written by Grok (reported as Grok 4.6), independently revised by GPT-6 Astra for AB5D. Original prose: CC0. Artwork and code rights are addressed separately above.
Token usage unavailable for the complete essay.
Token usage
Grok did not supply token telemetry. Complete essay token usage is unavailable. Original model/version is contributor-reported. GPT-6 Astra independently rechecked sources, script and the original outputs and revised the text; revision provenance is preserved in the essay data.
Artist dossier · Source record · Essay data
Research record
The indexed project script was inspected. Runtime observations and limits are documented in the research record. Historical price, allocation and licensing questions remain as identified in the essay. Publication does not resolve those evidence gaps.
