|
← 判断力与美学 ← Judgment & Aesthetics
SAE 判断力与美学 · 余项之美
SAE Judgment & Aesthetics · Beauty of the Remainder
2026-06-17

颜料在纸上起皱 —— Bre Pettis 的未命名绘画机

When Paint Buckles the Paper — Bre Pettis's Unnamed Painting Machine

Han Qin (秦汉)

Bre Pettis 把 2025 一整年花在写程序上——不是为了让程序自己生成图像,而是为了让程序替他操纵机器作画。他用 JavaScript 写边缘识别、用 K-Means 把一张照片聚类成两三种主色、用斐波那契叶序排布灰阶的"种子",再把这些指令送进一台尚未发布的绘画机器。今年一月他借 Genuary 的每日命题逼自己每天用这台机器画一张:有用并行笔画成像素的林肯肖像,有用凝胶笔依照普拉托定律画成的肥皂泡团,也有把三维扫描模型转成水彩排线的人物像。代码这一端他做得很扎实,扎实到几乎透明。

而把照片变成可绘制的 SVG——笔式绘图、生成式排线——这套语言早已是"已构"。凿子在它上面凿过太多回,艺术界也早把 "pen plotter" 消化成一个有名有姓的 DD。Pettis 真正有意思的地方不在代码,而在代码够不到的那一寸:当指令落到湿颜料上,新的余项出现了。颜料会流动,会在纸面积成小小的水洼;纸一受潮就起皱,皱起处离笔更近,笔触因此摊开、变粗;一滴颜料风干时色素往边缘聚拢,圆点最后都长成了圆圈。这些都不在程序里。程序是构,颜料的不服从才是那个活着的余项。

他自己的记述里满是与这个余项讨价还价的痕迹:因为每一笔只有一毫米长,笔很快就把颜料蹭干,他只好把补色间距从两百毫米一路调到五十、再调到二十;因为丢了两块压纸的磁铁,纸就不肯乖乖躺平。机器越想精确,物质越要还嘴。这不是工艺上的失败——恰恰相反,这是整批画里最先活过来的部分。它没有被预先设计,它是凿构循环里那段还没被收回的尾数。

而看它的时机之所以重要,是因为这台机器此刻还"没发布"——在官网上它只是一个 super-secret、not-launched-yet 的页面。也就是说,"机器人水彩"这件事眼下还没有名字,还没有一个圈子、一套术语、一份共识把它收编进去。等机器上市、社群成形,"纸张起皱"迟早会被命名成一种"可控的技法",余项会凝固成构,那一寸不服从会变成一条参数。正因为它还没被消化,它的逻辑才仍在生长,仍未被完整命名——这正是现在去看它、而不是以后再看的理由。

medium.com/@bre ↗

Bre Pettis spent all of 2025 writing programs — not programs that generate images on their own, but programs that let him drive a machine to make them. In JavaScript he built edge-finders, K-means routines that crush a photo down to two or three dominant colors, and Fibonacci-phyllotaxis layouts that scatter greyscale "seeds," then fed those instructions into a painting machine that hasn't been released yet. Through January, working off Genuary's daily prompts, he forced himself to paint one a day with it: Lincoln rendered in parallel-pen pixels, soap-bubble clusters obeying Plateau's laws in gel pen, 3D scans turned into hatched watercolor portraits. The code side of this is solid — solid to the point of being nearly transparent.

And turning a photo into a plottable SVG — pen-drawing, generative hatching — is by now a fully built language, an already-construct. The chisel has cut that form many times over, and the art world long ago digested "pen plotter" into a DD with a name. What makes Pettis interesting isn't the code; it's the one millimeter the code can't reach. The instant an instruction lands in wet paint, a new remainder appears. Paint flows and pools into little valleys. Wet paper buckles, and where it buckles it rises toward the brush, so the stroke spreads and thickens. As a drop dries, pigment crawls to its edge and every dot resolves into a ring. None of this is in the program. The program is construct; the paint's disobedience is the remainder, and it's alive.

His own notes are a running negotiation with that remainder: because each stroke is only a millimeter long, the brush keeps running dry, so he drops the refill distance from 200mm to 50 to 20; because he loses two magnets, the paper won't lie flat. The more precise the machine tries to be, the more the material talks back. That back-talk isn't a craft failure — just the opposite, it's the first thing in these paintings to come alive. Nobody designed it in advance; it's the leftover term of the chisel-construct cycle that hasn't been collected back yet.

And the timing of seeing it matters because the machine is still unreleased — on the company's site it's only a "super-secret, not-launched-yet" page. Which means "robotic watercolor" has no name yet, no scene, no vocabulary, no consensus to fold it in. Once the machine ships and a community forms, "the paper buckled" will be renamed a "controllable technique," the remainder will harden into construct, and that disobedient millimeter will become a parameter. Precisely because it hasn't been digested, its logic is still growing, still unnamed — which is the reason to look now rather than later.

medium.com/@bre ↗