Cushions

The work and its context

Cushions makes a grid appear capable of yielding. Its coloured regions can bend, swell or extend across the square like lengths of fabric, yet their edges remain sharply defined. Some outputs barely curve at all. In those works, the same underlying procedure produces angular corridors whose line endings resemble seams. The title provides a material association without requiring every image to illustrate a cushion.¹²³⁴

The series is a 200-work Art Blocks Factory release, project 231 on its Ethereum contract. Its official release date is 7 January 2022, following an earlier first token. The artist dossier supplies a broader practice context. Here the distinguishing problem is how a program can draw boundaries and then discover the shapes those boundaries have produced.⁵⁶⁷

The five examples examined below make that process visible in different ways. Broad bands invite the eye to follow one region across the field; finer grids make it harder to distinguish a single coloured body from the pattern of lines within it. Dark voids introduce another scale of activity through their small repeated motifs. The work's variety depends on these relationships as much as on its palette choices.¹²³⁴⁸

Conception and development

Parikh's own project account traces the core idea to 2018, when it was an interactive tool whose users selected parameters. Preparing it for Art Blocks transferred those decisions to token-derived variation. The account identifies a meeting between generative drawing and image processing: diagonal curves create the field, then connected-component analysis identifies the regions that become coloured forms.⁹

The development involved rejecting promising alternatives. Parikh describes composition experiments that weakened the desired softness, and matte palettes that moved away from the intended digital-illustration character. These choices explain why technical variation alone is not an adequate description of the work. The available differences were selected to preserve a recognisable visual family.⁹

The artist's project page credits Cushions to Devi Parikh and separately credits its webpage design to Abhishek Das. This agrees with the indexed artist name. The current Art Blocks interface attaches a combined artist-profile name, so this essay retains the project-specific credit rather than treating the profile grouping as evidence of joint authorship.⁹⁶⁵

How the work was constructed

Cushions begins with lines but finishes by finding the spaces between them. The inspected program uses p5.js 1.0.0. It splits the token hash into bytes for parameter selection and also uses its two halves to initialise two pseudorandom streams. Each stream is advanced repeatedly before drawing, and subsequent requests alternate between them. A fixed token therefore supplies a repeatable sequence of decisions, rather than requiring a fresh source of randomness whenever the image is opened.⁷

The first byte selects a grid density from fifteen stored values, ranging from four to sixty cells along an edge. Other bytes govern curvature, directional bias, line thickness, palette and the number of colours used from that palette. Several limits depend on density: fine grids receive narrower ranges for some of the parameters. The code therefore does not treat every visible feature as an independent switch. Its restrictions keep particular combinations within a designed range.⁷

In each grid cell, the program chooses one of the two diagonals. It draws that diagonal as a quadratic curve, with a displaced control point determining the bend. A probability controls the balance of diagonal directions across the field, while further seeded choices change which way individual curves bow. At one extreme, the field can look like straight, nearly parallel corridors. At another, short curling boundaries continually turn back into their neighbours.⁷

At this stage, the coloured cushions have not yet been drawn as separate objects. The program reads the canvas pixels and extracts their alpha values into an array. An embedded connected-component routine traces neighbouring pixels and assigns labels to contiguous regions between the drawn boundaries. The routine considers eight neighbour directions and explicitly closes the outer border. The process is a measurement of a raster image that the program has just made, rather than a catalogue of predefined cushion silhouettes.⁷

It then accumulates region sizes and positions. These measurements support approximate centres for the labelled regions, which select colours from a randomly populated eleven-by-eleven template. Large regions can consequently carry a common colour across several grid cells. A short curve may terminate within such a region without dividing it completely. That explains a distinctive visual effect in the examples: an internal dark line can seem to be a seam inside one coloured body rather than a boundary between two differently coloured bodies.⁷

Another branch sorts region sizes and selects large regions to remain as dark voids. The target fraction is chosen from three stored values, but the selection works with whole regions. It is not a command to darken an exact percentage of arbitrary pixels. The code also includes conditions for retaining or suppressing curves through these voids. A dark passage may therefore expose the underlying grid or appear as a more continuous interruption.⁷

The program draws its ground and optional accent pattern before replacing the pixels of the selected coloured regions. Accent routines include dots, circles, squares, crossed lines and several curved motifs. Because the coloured regions cover this preliminary pattern, it remains visible principally in the unfilled spaces. The drawing order gives the apparent texture a specific role: it belongs to the dark interval between coloured areas rather than being indiscriminately spread over the whole image.⁷

Finally, the stored curves are redrawn. One mode adds a narrower light line inside a thicker dark boundary; another retains a single dark stroke. This final pass reconnects the raster colouring to the initial geometry and makes the borders crisp. A frame closes the square. The combination of sampled region analysis and redrawn curves accounts for the work's simultaneous softness and precision without requiring a simulation of stuffed fabric.⁷

The script calls noLoop in setup and produces a static drawing. No audio or interaction handler was identified. For ordinary viewport sizes it adjusts pixel density toward a 2,400-pixel backing image, making the distinction between display size and the pixel array used for region analysis consequential. The five originals were inspected in a desktop browser; cross-browser pixel equivalence and behaviour at other output dimensions were not established in this review.⁷

The mint

The first indexed token was minted on 12 December 2021 at 17:38:02. Activation is recorded on 17 December at 01:52:45. The scheduled start is 7 January 2022 at 20:00:00, with the second token minted forty-four seconds later. The collection page gives 7 January as the release date. The earlier first token and the public release therefore concern different events.⁶⁵

Completion is recorded at 20:06:38 on 7 January. This is six minutes and thirty-eight seconds after the scheduled start, or five minutes and fifty-four seconds after the second mint. Those are intervals between named index fields. Without classifying the underlying transactions, they do not establish a precise uninterrupted public sellout duration or identify the purpose of the early token.⁶

The current record contains 200 invocations and a maximum of 200. Its sales notes say fixed price, and the configured price is 0.15 ETH. The configuration was retrieved during this research; original announcements and transaction receipts would be needed to establish historical price changes or the amount paid for each token. Current edition completion does not independently establish every term of the original offer.⁶

For the header illustration, token #150 was checked against the complete set of three wallets linked to the AB5D profile. Its indexed owner matches that set. This is an ownership observation for the displayed work, separate from the release chronology. The original query and full timestamp offsets are preserved in the research record.⁶⁸

Copyright and licensing

The indexed artwork label is CC BY-NC 4.0. The corresponding Creative Commons deed allows sharing and adaptation with attribution, an indication of changes and a noncommercial condition. The project record provides no separate terms link, and this review has not reconstructed earlier licence settings. The caption gives the observed label without treating token ownership as a transfer of copyright.⁶¹⁰

Two dependency records need to remain distinct. The p5.js 1.0.0 repository carries LGPL 2.1. The artist script also ends with a reference to bramp's connected-component-labelling implementation and an Apache 2.0 notice; the upstream repository supplies that licence. These grants concern their respective software and do not establish a separate licence for Parikh's surrounding program. No independent artist-code grant was identified in the inspected material. The essay's original prose is dedicated to CC0, with artwork and software rights kept separate.⁷¹¹¹²

Close reading of five outputs

The comparison uses editions 0, 1, 10 and 20, selected as an early-edition group, plus the collection-held #150. Each official generator was opened and visually inspected. This is a small comparison chosen for close looking, not a random sample from which to estimate the frequency of palettes, voids or grid settings.¹²³⁴⁸

In #0, cream and warm brown regions run diagonally through a black framework. A narrow light line sits inside the darker outlines, making the borders appear layered. The broad brown passage through the centre changes direction and contains a long line that ends before it cuts the region apart. That termination matters: it asks the viewer to read the shape as one connected body with an internal division, rather than as a row of independent tiles.¹

A branching black region crosses the upper and middle portions of the image. Small repeated curves texture its interior, introducing a finely spaced rhythm beside the large flat colour fields. The contrast makes the black area feel simultaneously recessed and busy. It is not an absence of drawing. Following the small motif briefly distracts from the larger diagonal route, after which the light-edged borders restore the broad structure. The resulting shifts of scale are more interesting than simply identifying the work as a textile-like pattern.¹

Edition #1 removes much of that contrast in finish. Ochre-orange dominates, while terracotta occupies a broad area toward the lower left and smaller fragments near the edges. Thin dark straight lines divide the field into long angular passages. Several stop within a shared colour, leaving the surrounding region continuous. There is no patterned dark channel comparable to #0. The image consequently feels flatter and more economical, with colour and line doing almost all of the work.²

The dominant diagonal is strong enough to suggest a regular layout, but local turns interrupt it. An orange corridor can be followed for a long stretch before an angled return sends the eye elsewhere. Because the line is thin, the coloured region retains more visual weight than the boundary. Compared with #0, there is less suggestion of raised edging or decorated fabric. The same connected-region logic can support a comparatively spare arrangement of paths.²

Edition #10 uses pink and pale cyan to build a field of wavy diagonal bands. Their dark outlines repeatedly swell and narrow, producing a soft lateral pulse. One broad dark channel crosses from the upper-left area toward the lower right. Inside it, tiny outlined squares repeat on a regular grid in the two principal colours. The squared motif makes the surrounding waves seem more fluid by comparison. The channel also interrupts what would otherwise be a nearly continuous rhythm of alternating coloured strips.³

Edition #20 enlarges the curve. Pink, deep blue, mint and pale yellow form broad scalloped regions, with far fewer turns across the image than in #150. A central dark passage carries rows of coloured dots. Its large bends seem to pass between soft lobes, but the image gives no actual shading or fabric texture to support a literal object. The material suggestion comes from the contour. Short boundaries within the broad blue area remain visible, demonstrating again that a drawn segment need not divide a region into separate colours.⁴

The AB5D example, #150, compresses the vocabulary into a dense field of pink, purple and turquoise. Dark curling boundaries fill almost every part of the square. At first the eye may treat these as many small independent forms; sustained looking reveals broad connected patches of colour extending through them. A large pink area near the upper middle and turquoise passages toward the right create a second, coarser organisation beneath the fine linework. There is no large patterned void to provide a rest.⁸

That density changes the viewing distance at which the composition becomes legible. From farther away, the three colours arrange themselves as irregular territories. Closer inspection reveals the repeated turns that organise each territory internally. The contrast with #20 is substantial: large scallops make the boundary of an individual region immediately prominent, while #150 makes the viewer work to separate a region from the intricate marks it contains. These are differences of structure as well as colour.⁴⁸

Significance and limitations

Cushions gains its identity from a reversal of emphasis. The program starts by drawing a network, then treats the spaces inside that network as the material of the final image. Its connected-component analysis is not merely an invisible efficiency measure. It is the operation that allows a region to cross several cells and acquire an identity larger than the grid from which it arose.⁷

The examples make that identity perceptually unstable in a productive way. In #1, a line ending remains a precise graphic mark inside an expansive coloured passage. In #150, many such marks compete with the larger colour territories. In #10 and #20, patterned voids introduce a different internal scale. My reading is that the work's softness comes from this negotiation between boundaries and continuous regions, rather than from any attempt to reproduce the physical behaviour of a cushion.²³⁴⁸

Unresolved research questions:

• Recover original sale announcements, price history and allocation records.
• Compare the indexed artist script with a pinned on-chain retrieval.
• Identify any separate artist-code reuse grant and earlier artwork terms.
• Test raster labelling and rendering equivalence across browser versions and output dimensions.
• Extend the visual comparison beyond the five inspected editions.