LeWitt Generator Generator

A drawing and the wall beneath it

LeWitt Generator Generator #563 is the AB5D collection example for this essay. The original renderer presents a dense blue network above a yellow-to-white dotted surface against red. Two observations showed the same completed image. This is a static work generated in the browser, not an animation waiting to begin.¹²

Mitchell F. Chan released LeWitt Generator Generator through Art Blocks Factory on 4 August 2021. It is Ethereum project 118 on contract 0xa7d8d9ef8d8ce8992df33d8b8cf4aebabd5bd270; 750 tokens are currently indexed. Its stated reference is Sol LeWitt's Wall Drawing #118, reimagined on algorithmically constructed, non-flat walls.³

The duplication in the title names a real formal distinction. Chan generates a drawing procedure's support as well as its marks. The image keeps both visible: a field of regularly arranged dots describes the wall, while a separate web traces connections across it. Those layers can be compared, so the viewer sees the conditions of a drawing as part of the drawing's subject.⁴¹

A return to an earlier argument

Chan's essay dated 26 July 2021 gives a direct account of intention. He frames NFTs through a distinction between an artwork's expressive form and its commodity form. He also identifies this project as a reconsideration of his 2014 Sol LeWitt Generator at Angell Gallery: there he distributed code instructions freely, wrote code on the wall and sold generated prints. In his retrospective account, the earlier work had expressed doubt that freely shareable ideas could sustain the conceptualist model of artistic value; the 2021 project revisited that conclusion.⁵

This is an artist's argument about ownership and artistic form. It should not be mistaken for a licence granting a token holder copyright. Its importance here is that the reference to LeWitt is documented intention, not a similarity inferred from the title alone.⁵

The historical point of departure is also specific. Fundación Botín's exhibition checklist identifies Wall Drawing 118 as fifty randomly positioned points joined by straight lines, first installed at the School of the Museum of Fine Arts Gallery in Boston in December 1971. It names the initial draughtspeople, making execution by others part of the record.⁶

Chan's own project page stages the transformation: establish a wall, distribute points, connect them, change the viewing angle and distort the support using one of six algorithms. That sequence is an explanation of how to imagine the work. The final script, discussed below, adds its own decisions and does not enforce fifty points in every output.⁷

Constructing a support and a network

The full 16,842-byte artist program was retrieved from the official project index and read. It uses p5@1.0.0 with a WEBGL canvas. Hash bytes map directly to parameters for palette, distortion family and depth, backdrop, wall proportion and view angles. A separate numerical seed derived from the hash prefix drives a small xorshift generator, and the same seed initialises p5 noise. This separates stable edition parameters from the sequence of choices placing points.⁴

The program's point count is a revealing qualification to its public explanation. Although a placement helper is named placeFiftyPointsAtRandom, setup chooses an integer from 12 through 51. Each point receives two coordinates in a square domain. Every distinct pair is connected once. For n points the result is n(n−1)/2 connections, so an increase in points produces a much faster increase in paths. Density is therefore a structural choice, not merely a heavier stroke applied to the same drawing.⁴

Each connection begins as a straight segment in the flat coordinate plane. The program samples it at 25 positions, divides it into 24 intervals and assigns a height to each sampled position. Six alternatives determine that height: a mound, two mounds, waves, an egg-carton pattern, a saddle or seeded noise. Trigonometric functions create the regular undulations, a difference of squares produces the saddle, and noise supplies the irregular surface. Curves interpolate the displaced samples when the connection is drawn.⁴

Consequently, the finished paths need not look straight on screen. Their straightness belongs to the initial planar relationship between endpoints; the transformation bends their intermediate positions. The work makes that distinction inspectable because the dotted wall uses the corresponding height rule. A smooth hill and a line rising across it are two expressions of the same transformation. The image can look sculptural even though neither layer becomes a solid rendered object.⁴

Chan deliberately separates the layers along the depth axis. The wall is translated backward and the connection network forward by equal distances, each one-eighth of the smaller canvas dimension. The scene is then rotated about two axes and viewed with an orthographic projection. The foreground web therefore floats away from the surface that explains it. This separation creates space for comparison, rather than making the dotted support disappear under the drawing.⁴

Colour is likewise organised by role. Ten stored palettes provide values for background, backdrop, network and wall. The wall interpolates between two colours across one coordinate, while connection colours interpolate according to their target node. Line thickness changes in broad bands according to point count: fewer points receive heavier strokes. Backdrop selection can introduce a square, diamond or circle into a wider gradient field. These are deliberate compositional constraints surrounding the variable geometry.⁴

All visible drawing happens in setup; draw is empty. No mouse, keyboard or touch interaction handlers, audio calls or external image loads were found. The held original remained unchanged across two screenshots, consistent with that construction. The browser still needs the specified p5 runtime and WebGL support. Exact cross-browser pixel equivalence was not tested, and no explicit post-load window-resize handler was located.⁴¹

The launch and the earlier token

The earliest indexed mint is 25 July 2021 at 20:12:07 UTC, corresponding to #0. Activation follows on 27 July at 17:00:39. Neither date is the public release date shown by the collection page. Chan's dated essay announces minting for 4 August at 11 a.m. CDT, equivalent to 16:00 UTC.³⁵⁸

The second indexed mint occurred at 15:59:36 UTC on 4 August; the index's start field matches it. Completion is recorded at 16:07:07, an interval of seven minutes and 31 seconds from that second mint. The 24-second difference between the announced hour and the observed second mint is retained rather than rounded away. These records establish a tightly concentrated launch around the advertised time, with one substantially earlier token. They do not establish the purpose or recipient category of that first mint.³⁹

The current index reports 750 invocations, a maximum of 750 and a stored price of 0.75 ETH. Those are present record values. They do not prove an unchanged maximum from inception, a uniform historical purchase price or the absence of reserves. An archived sale configuration and a full classification of mint transactions would be needed to answer those questions confidently.³

The recorded licence and its limits

The project record reports NIFTY License but supplies no link to operative terms. That label is preserved exactly. It is not treated as public-domain status, a Creative Commons licence or an unconditional right to reproduce the work. The artist's project site also carries his copyright notice.³⁷

Dapper Labs' public NFT License page provides a relevant comparison, including version 2.0, revised 5 November 2018. Its terms retain creator ownership, distinguish personal use from specified commercial merchandise rights and condition those grants on continued token ownership. They also contain restrictions on modification and conditions for third-party website display. However, the inspected project record does not identify that version or a project-specific adoption text. The generic document is therefore useful for understanding what requires verification, not proof that every provision applies to this edition.¹⁰

For #563, this package records the original hosted asset, token link, current owner and exact rights label. Public editorial display should retain Chan's credit; the basis for that display remains an explicit review item rather than an invented permission inferred from ownership. Other examples are discussed through links and original observation, with no copied artist program included.

No separate licence for Chan's code was located. The p5.js 1.0.0 dependency has LGPL 2.1 terms, which govern the library rather than granting rights in the original artwork or artist script. Likewise, the documented reference to LeWitt establishes a conceptual relationship; this research does not assert an endorsement or permissions agreement from LeWitt's estate.⁴¹¹

Art Blocks’ June 2021 licensing guide identifies NIFTY as shorthand for NFT License 2.0. This strengthens the version attribution, although project-specific adoption and historical changes remain unverified.¹²

Four different relationships to a wall

The selected editions are #0, #1, #10 and #20, each verified against the exact contract and viewed in its original renderer. They are an early-edition comparison, not a frequency study. The examples show how point density, surface shape, view and colour can make the same construction read as fabric, a cage or a drawn landscape.⁸⁹¹³¹⁴

In #0, a green network rises into two broad crests above a paler dotted sheet. Its points gather into sharp local tips, but the paths between them bow and flow. Beneath and behind it, the wall extends far beyond the web, with blue-grey dots describing repeated swells. The softly coloured background contains a faint square that gives this apparently free-floating form a second frame. The indexed attributes name Jade GC, Twin Peaks, Medium distortion, Moderate complexity and a Square backdrop.⁸

The significant visual tension is between local convergence and continuous surface. Where many lines meet, the network looks pulled into a knot; away from those junctions it resembles fine threads draped over a hill. The dotted layer changes that reading. Because its samples remain regularly spaced in their original domain, the viewer can compare an ordered surface with the irregular locations chosen for the foreground points. The result feels suspended between a landscape diagram and an unsupported textile. Those are interpretive comparisons, not identified subjects supplied by the metadata.⁸

Edition #1 is more open and immediately legible as a network. Thick dark-purple paths form a shallow bowl-like arrangement against a peach-to-yellow field. Several widely separated upper nodes send fans of curves toward lower junctions. The pale wall behind it becomes almost atmospheric where its dots approach the background colour, but brighter ridges still disclose its shape. Its recorded combination is Joker, Bunker, CALM distortion, Elegant complexity and Clear backdrop.⁹

Compared with #0, this example allows individual connections to be followed for longer before they merge into a dense texture. The code's density-sensitive stroke weights help explain that difference without turning the adjective Elegant into an objective judgement of quality. The lower pocket appears to hold the lines, although it is only their shared displacement that produces the illusion. This is a persuasive demonstration of how a rule about connecting points can acquire the visual suggestion of gravity without modelling cloth or physical tension.⁴⁹

In #10, the green foreground forms a dramatically unequal bow-tie silhouette: a tall, densely connected left side narrows towards a central crossing, while a smaller right lobe opens outward. The wall below repeats a related narrowing in yellow and magenta. Against a warm pale background, the two layers make their separation especially clear. The index identifies Carolina Will, Saddle, Medium-Calm distortion and Challenging complexity. The saddle's mathematical opposition is visible as a change in direction, rather than simply as increased roughness.¹³

Edition #20 returns to Twin Peaks with a very different hierarchy. A brilliant white web stands out against grey; the dotted wall is dark and recedes behind it. Dense overlaps make some regions almost luminous, while the edges resolve into individual curving paths. Metaregular colour, Medium distortion and Moderate complexity are indexed with a Clear backdrop. Comparing it with #0 shows why surface family alone does not describe an output: contrast can reverse which layer commands attention even when the geometric category agrees.¹⁴

What the second generator makes visible

The series is strongest when viewed as a set of relations, rather than as 750 illustrations of one historical reference. Support and inscription share a formula but occupy different depths. A fixed rule for joining points produces different densities. A static result retains the evidence of its transformations without animating them. The program's variability makes those distinctions available across editions.⁴⁸¹⁴

The historical wall drawing and Chan's software need not be treated as interchangeable to appreciate the connection. The former's fifty-point instruction is a specific documented precedent; the latter adds variable counts, virtual depth, coloured sampling and a generated support. Those changes make Chan's authorship visible at the level of the rules, as well as in the resulting images.⁴⁶

Unresolved research questions:

• What were the exact opening price, original maximum and reserve arrangements, and what was the purpose of the July #0 mint?
• Which operative NIFTY licence version or project-specific text was adopted, and what establishes this archive's display permission?
• Was any separate licence granted for the artist script or any agreement documented concerning the historical reference?
• How were the palette names and six distortion families developed and selected?
• How consistent are WebGL line and point rendering across browsers, display densities and resized windows?