AI 艺术与普通艺术的区别（2）： AI 绘画的彩色语言与点彩主义绘画的色彩语言之比较

文：施舒

点彩画派与它的创始人修拉

只要谈到后印象派绘画，法国艺术家乔治·皮埃尔·修拉就是一个绕不过去的主题。修拉是后印象主义运动的发起者。整个著名的点彩派绘画风格（也狭义地被称为色度主义），这面后印象主义运动的旗帜，原则上都是由修拉单枪匹马一手设计的。

截至修拉所生活的年代为止，修拉是历史上为数极少的几位真正肯花大量时间研究色彩理论的画家之一。他将自己的研究成果应用于绘画实践，创立了点彩画派。 他的驰名作品，如《大碗岛上的一个星期日下午》和《骚动》，改变了现代艺术的走向，成为19世纪晚期绘画带标志性的一道风景线。

修拉的点彩主义（以及从中发展起来的，涉及面更为广泛的后印象主义）的形成，深受当时法国科学界的色彩理论的影响。当时法国的色彩理论的主要代表是法国化学家雪佛勒（M.E. Chevreul）。雪佛勒发现，如果将两种颜色并列在一起，从远处看时，它们就可以在人的眼睛里"合成"出另一种颜色。 他还注意到，盯住某个颜色看一段时间，便会在观察者的眼睛中"产生"一种"光环"。这种"光环"的颜色通常就是与原本的那种颜色“相反”的颜色，在色轮上这种“相反”的颜色也被称为色相对比色（或称补色）。如一个人长时间地凝视一个红色的物体后，眼中就可能会在红色的物体之外看到青色的光环。这是一种幻觉。事实上并没有一束青色的环状光照射在那位观察者的眼中。后来人们知道，这种虚幻的对比色的产生可以用视网膜的视觉信号的滞留来解释。

鲁德（O. Rood）也对点彩主义的形成产生过影响。 鲁德指出，将原色（如红色、绿色和蓝紫色）的色块或色点彼此并列，能创造出比相应颜料的物理混合更强烈、更赏心悦目的颜色来。 在点彩主义诞生之前，世上的画家们并不知道这一点。画家们只知道通过混合和搅拌两种或两种以上的颜料来制造“新的”颜色（混合色）。点彩主义诞生之后，一种全新的色彩方法出现了。

鲁德还指出：对颜料进行“加法”或“减法”处理后，得到新颜色之间常存在着色泽品质的差异。这是因为颜料的混合方式和光的混合方式是不同的：

· 颜料：红色 + 黄色 + 蓝色 = 黑色

· 光：红色 + 绿色 + 蓝色 = 白色

修拉把色彩理论的这些重要结论都记在了心里，并对色彩理论家按物理学的原理去理解和解释绘画现象的方法深信不疑。 他坚信，就像音乐家可以借用音符和曲调的组合和变异来创造一维的和谐一样，画家可以用色彩的组合和变异在画布上创造出二维的和谐。 修拉相信，十九世纪的色彩理论就是一条普适定律。他下决心通过自己的实践去证明：凭借对视觉感知和光学定律的理解，就可以创造出一种全新的艺术语言。 于是，他开始使用各种线条、色彩强度和配色方案来创造这种艺术语言。 修拉称这种语言为"色度主义"，此即后来的，也更加广为人知的 "点彩主义"。点彩主义不仅在在视觉艺术界，而且在音乐界和文学界的都引起了巨大的轰动和反响，最终发展成了跨越诸多文化领域的一场被称为"后印象主义"的文化运动。

点彩画派的理论精髓可以归纳如下：

1. 让耀眼的暖色调占据画面的主导地位，以及使用向上方运动的线条，都可以用来给画面注入喜悦的情绪。

2. 取得明色和暗色的平衡、暖色与冷色的平衡，以及使用水平运动的线条，都可以用来的来维持画面情绪的平静。

3. 深色和冷色，以及向下运动的线条，可以用来强化画面的抑郁情绪。

而点彩主义色彩语言的关键词汇则包括：色点、并置、主导、色调、视觉融合、色彩情绪。

在19世纪的艺术界，点彩主义将油画从一门手工艺技升华到了一门可以由物理学理论来指导的学问。在19世纪至21世纪的这样很长的一段时间里，点彩主义的诞生一直被文化精英们视为一场艺术革命，因而深受推崇。

例如，以修拉的《大碗岛上的一个星期日下午》为素材和灵感源泉，斯蒂芬·桑德海姆和詹姆斯·拉帕茵创作出了著名的音乐剧《星期天和乔治一道在公园里》。后来詹姆斯·拉帕茵又以此为题写了的一本书。

这部音乐剧一战成名，连续获得1985年普利策戏剧奖、1985年戏剧台优秀音乐剧奖、1985年戏剧台优秀图书奖、1985年戏剧台优秀歌词奖、1991年劳伦斯·奥利维尔最佳新音乐剧奖、2007年奥利维尔优秀音乐剧奖等。

点彩派绘画与 AI 绘画的比较

无疑，修拉对所有这些荣誉都当之无愧。从19 - 20世纪绘画工具和绘画实践的大背景下去看，尤其如此。 然而，在AI时代的今天，在各种类型的计算工具和人工智能唾手可得的新情况下，作为新一代艺术家，我们是否应该自问：我们怎样才能比前人（或比过去的大师们）做得更好？

图1是一幅AI 绘画，题为《API0029 - 驯服怪兽：向修拉的《骚动》致敬》，由施舒与他的智能虚拟助手ESAG共同完成。(图1来源： https://www.esandag-ai-art-studio.com)。 图2是著名的点彩派绘画《骚动》，作者：修拉。（图2来源：https://en.wikipedia.org/wiki/Le_Chahut）。

比较图1和图2可以看出 ，AI绘画至少在以下几个方面更为得心应手：

(1) 在部署颜色（色调）梯度方面：虽然图1和图2的画家，使用了相似的绘画语言来表达的主题，各自都选择了使用色块来构图。例如，各自都选择了使用色块来描述画中人物的帽子，面孔，颈部和手臂，以及服装。但图1的颜色层次更多、更深、也更富于变化。因此，相较于图2，图1看上去就更加流畅和更有感染力。

这种流畅和感染力直接来自于颜色（色调）梯度的建立。在建立二维梯度阵列（即梯度场），确定梯度阵列的定量细节（例如，各局部颜色强度最大值的坐标位置；各局部颜色强度最小值的坐标位置；在沿着平面360度角中所有需要考虑的方向上，颜色强度从局部最小值变到局部最大值时，变化速度的快慢或的剧烈程度的大小；抑或反过来，颜色强度从局部最大值变到局部最小值时，变化速度的快慢或的剧烈程度的大小），以及从上述所有可能的排列组合中筛选出最优的方案等方面，与人脑相比，AI明显具有记忆和运算方面的优势。

从梯度场的角度看，图2所使用的梯度场相对比较原始和粗糙：人物腿部（或丝袜）的阴影、乐器上的阴影、以及音乐指挥（图2左下角）西装上的阴影，均由深棕色衬以浅棕色来刻画。虽然在刻画墙上灯罩的阴影时，画家尝试了复杂一些的颜色梯度，例如将深蓝色与不同灰度的深棕色并列点彩。但总体而言，画面的颜色依然显得沉闷而迟钝。相比之下，如图 1 所示，AI在解决起这类问题时，却轻车熟路、游刃有余，可以组织将颜色组织得更加生动活泼，富于想象力。

(2) 在设计和处理颜色块边界二侧的梯度符号（即二次导数场）方面：当两个色块（或色点），例如 R（红）色和 B（蓝）色，彼此相邻时，下述情况便会出现：

a. R色和 B色的色彩强度都随着接近其共同的边界而上升。 我们将此称为 R+ 对B+ 的情形。（此处“+”号来自于色彩强度的二级偏导数均为正数。）

b. R色和 B 色的色彩强度都随着接近其共同的边界而下降。 我们将此称为 R- 对B- 的情形。（此处“-”号来自于色彩强度的二级偏导数均为负数。）

c. R色接近边界时色彩强度上升，而 B 色接近同一边界线时色彩强度下降。 我们称之为R+ 对B- 的情形。

d. R色接近边界时色彩强度降低，而 B色接近同一边界线时色彩强度上升。 我们称之为 R- 对B+ 的情形。

依据R色与B色在色轮中的相对位置，R色与B色的相对亮度（即灰色），以及R色和B色在"温度"标尺上的相对位置（如中等暖色对极度冷色）的不同，上述4种情形通常会给看画的观众留下不同的情绪印象。例如，某些情况下画面显得冲突、另一些情况下则显得和谐、有一些情况下显得清晰、另一些情况下则显得混乱、有一些情况下显得活泼、另一些情况下则显得沉闷等。

AI引入到颜色语言中的新词汇包括：色彩梯度场、矢量色彩梯度二维阵列、梯度符号、二 阶矢量色彩梯度和二阶色彩梯度场。

一幅画，经常需要涉及几十种颜色，数百甚至上千个彼此相邻的颜色对。要记住所有这些细节并同时对涉及这上千个颜色对的梯度场进行优化，确实是一个人类大脑力所不及的任务。然而，对于一款设计正确的AI来说这却是一件轻而易举的事情。一个耗费了人类顶级画家数年时间才完成的绘画任务，交给AI去做，1 – 2个小时之内，便可轻松地完成。不仅能完成，而且能够画得更为出色动人。 AI很轻松地就可以计算清楚如何避免某一部位的和谐被另一部位的冲突所抵消，或某一部位的活泼被另一部位的沉闷所冲淡。

不仅比较图1和图2可以佐证这一点。比较图3和图4也可以佐证这一点。图3是修拉创作的《大碗岛上的一个星期日下午》

（图3来源：https://en.wikipedia.org/wiki/A_Sunday_Afternoon_on_the_Island_of_La_Grande_Jatte）。

这幅作品是公认的点彩派绘画的旗帜和最高成就。这幅画耗费了修拉2年的时间。期间，他绘制了60余幅研究性的油画稿，然后才正式发表。图4是一幅AI 绘画，题为《API0048 巴塞罗那黄昏的色彩 • AI 视角》，由施舒与他的智能虚拟助手ESAG共同完成，作画耗时不及2小时。

(图4来源： https://www.esandag-ai-art-studio.com)。

在此，我们不妨重新审视一下本文第6段中列举的点彩派绘画理论的要点之一：使用光亮的暖色调可以诱导出喜悦情绪。这个结论对不对？当然是对的！ 然而，借助比较图3和图4，不难看出：AI 为我们指出了一个超越前人，创造出更多美感的新方法、新机遇。如果我们停留于点彩派绘画理论，我们便会错失这一大好机遇。

长期以来，《大碗岛上的一个星期日下午》一直作为美术教科书中的典型教例，被用来阐述点彩派的绘画理论。例如，图3用暖色覆盖了画布中、上部的大块面积（参图中的绿黄色和粉红色），并且采用了将大块暖色与大块冷色直接并列（即不使用过渡色，而是让阳光照耀的草坪和阳光照不到的阴影区沿着尖锐的边界线直接并置）的方法，通过强化冷暖对比来加强画面的明快感。

无疑，这幅画确实给人一种明亮和欢快的感觉。尤其是，如果我们站在19-20世纪的视觉艺术的角度看问题，那就更加毋庸置疑了。然而，我们已经进入到了在21世纪的AI时代。从这个新的角度看问题，尤其是如果我们将图3与图4相比，我们便不禁意识到，图3并没有找到提高画面的明快感的最佳方法。

借用AI，我们就并不一定需要使用大面积的暖色。有时只需较小面积的暖色块即可（如较小的黄色块，绿色块，和红色块）。也无需采用很强烈的对比度，只需使用设计正确的色彩梯度场以及正确的二阶梯度场，便可以在一个更高的层次上创造出快乐的情绪。图4可以说明这一点。

结论

1. 在点彩派已有的绘画理论中，如果再引进色彩梯度场和梯度符号（即二次导数场）等概念，其有效性和丰富程度便可得到充分的扩展。 这可以帮助增强作品的情绪表达力和色彩感染力。

2. 借助数字语言思维的AI与借助（定性的和相对模糊的）自然语言思维的人脑，天然是可以互相补充、相互增强的。 AI可以拨开数值运算的重重迷雾，帮助我们人类看到在过去几千年中我们从来未曾看到的各种崭新的可能和机遇。 对于我们这些生活在AI时代的黎明中的人类艺术家来说，不应再对AI的巨大潜力视而不见，或继续将时间和精力浪费在与AI打擂台上了。 我们应该自问的是：我们如何才能更好地与AI携手合作，创造更美好的未来？

AI Arts vs. Conventional Arts (2): Color Language of AI Paintings vs. That of Pointillism

By Eric S. Shi

Pointillism and G. Seurat

When it comes to post-impressionism art, Georges-Pierre Seurat, the French artist, and the initiator of the post-impressionism movement, is probably the most prominent corner stone of the grandeur monument. The famous pointillistic painting style (also narrowly known as chromoluminarism), the flag of the post-impressionism movement, was devised by Seurat, single handedly.

Seurat was, up to his time, one of the very few painters who spent a lot of time in studying color theories and applying them to his painting practice. His innovative works such as, “A Sunday Afternoon on the Island of La Grande Jatte” and “Le Chahut”, altered the direction of modern art, and collectively became one of the icons of late 19th-century painting.

The formation of Seurat’s pointillism (and the broader post-impressionism developed from it) was probably most influenced by the color theory developed by M. E. Chevreul, a French chemist. Chevreul discovered that if two colors juxtaposed very closely, they could “synthesize” another color in human eyes when seen from a distance. Chevreul also recognized that a "halo" color can be “produced” after staring at a color for a while, where the “halo” color is the opposing color (also known as contrast color and/or complimentary color) in the color wheel. For example: after looking at a red object, one may see a cyan echo/halo color around the original object. This imaginary formation of the contrast color is later explained as a result of retinal persistence.

O. Rood also exerted influence on the formation of pointillism. Rood pointed out that the juxtaposition of primary hues (e.g., red, green, and blue-violet) next to each other would create a far more intense and far more pleasing color, when perceived by the human eyes, than the physical mixing of corresponding color pigments would. Before pointillism, painters only knew the latter method, i.e., to create a color (or a hue) by mixing two or more other colors (or hues).

Rood also pointed out the difference between additive and subtractive qualities of color, since pigments and light do not mix in the same way:

· Pigments: Red + Yellow + Blue = Black

· Light: Red + Green + Blue = White

Seurat took these color theories to his heart, and came to believe the color theorists' notions on scientific approaches to paintings. He believed that, just like a musician could use counterpoint and variation to create 1D harmony in music, a painter could create 2D harmony on canvas, using hues and shades. Seurat came to believe that color theories at his time were like a universal law, and he was driven to prove that the knowledge of perception and optical laws could be used to create a new language of art based on its own set of heuristics and he set out to show this language using lines, color intensity and color schemes. Seurat called this language chromoluminarism, later more popularly known as pointillism which in turn initiated a movement known as post-impressionism across the whole horizon of visual art and music and literature.

The essence of pointillism theories includes the following:

1. The emotion of joy can be liberated by the establishment of predominance of luminous hues (i.e., the warm colors) and by the use of lines pointing upward.

2. Calm can be established by balanced deployment of light color vs. dark color, warm color vs. cold color, and lines that are horizontal.

3. Sadness can be enhanced by using dark and cold colors and by lines directed downward.

The key vocabularies of the pointillistic color language include: juxtapose, color dot, predominance, hues, visual blending, emotion.

In the art world of 19th century, the pointillism represented a major paradigm shift, liberating oil painting (which was nearly synonymous to art) from a handcraft skill to something can be guided by physical science. It was very much admired as a revolution in art by elites of society in a wide span of time, of 19th – 21st centuries.

For instance, Seurat’s “A Sunday Afternoon on the Island of La Grande Jatte” was so stimulating, it inspired Stephen Sondheim and James Lapine to produce the famous musical, “Sunday in the Park with George” and later a book by James Lapine.

The musical was so successful, it won 1985 Pulitzer Prize for Drama, 1985 Drama Desk Outstanding Musical, 1985 Drama Desk Outstanding Book, 1985 Drama Desk Outstanding Lyrics, 1991 Laurence Olivier Award for Best New Musical, and 2007 Olivier Outstanding Musical, among others.

Compare pointillistic paintings with AI paintings

The fact that both the pointillistic and AI paintings are constituted of color dots makes their comparison sensible. Clearly, Seurat deserves all the above-mentioned honors, and much more, especially in the context of painting practices and tools of 19th – 20th centuries. However, as the artists living in the dawn of an AI age, with various types of computer tools and capabilities available, does it make sense to question ourselves on: how can we do better than the past masters? AI augmented painting is, at least, one of the possible answers. The comparison below, hopeful, will demonstrate that AI painting does have certain merits and hold promises.

Figures 1 and 2 compare an AI painting “API0029 《Taming the Monster, Saluting Seurat’s Le Chahut》” by ESAG (short for Eric Shi Art Generator) (Source of Figure 1: https://www.esandag-ai-art-studio.com) and a famous pointillist painting “Le Chahut” by G. Seurat (Source of Figure 2: https://en.wikipedia.org/wiki/Le_Chahut).

Figure 1 Figure 2

From the comparison, it is clear that AI painting holds promises at least in the following areas:

(1) In deploying color (hue) gradients: Although both painters, on Figure 1 and Figure 2, approached their respective themes similarly and built corresponding images with blocks of color dots, e.g., in hats, in faces, necks, and arms, and in costumes, the colors in Figure 1 command more depths and variations and are consequently more fluid and emotional.

Such fluidity and sensation are a direct consequence of establishing color gradient in the AI painting. AI is advantageous, as compared to human brains, in calculating complex 2D arrays of gradients, sorting out quantitative details of the gradient fields (e.g., where the local maximums and/or local minimums of color intensities should be positioned, and how sharp the changes in vectorial gradients should be implemented going from a local minimum to a local maximum, or vice versa), and then alternating all these in an iterative way in order to optimize.

To this end, the shadows on legs (or stockings), music instruments, and the suit worn the music conductor (to the lower left corner) in Figure 2 are plainly colored in darker brown against less dark brown. Although efforts were made in setting up sorts of color gradients in painting shadows of lampshades on the wall where dark blue juxtaposed with dark brown of different grey scales were employed, the colors remained dull and obtuse. Comparatively, when AI is tasked to resolve a similar shadow problem, the colors came out more vivid and more imaginative, as manifested in Figure 1.

(2) In handling signs of the gradients across borderlines of the color blocks: When two color blocks (or dots), say R (for red) and B (for blue), are placed adjacent to each other, a number of scenarios may emerge:

a. Both R and B approach their common borderline with increasing color intensity. We call this an R+ vs. B+ scenario. (The plus sign comes from the secondary differential of the color intensity being positive.)

b. Both R and B approach their common borderline with decreasing color intensity. We call this an R- vs. B- scenario. (The minus sign comes from the secondary differential of the color intensity being negative.)

c. R approaches the borderline with increasing color intensity whereas B approaches the same borderline with decreasing color intensity. We call this an R+ vs. B- scenario.

d. R approaches the borderline with decreasing color intensity whereas B approaches the same borderline with increasing color intensity. We call this an R- vs. B+ scenario.

Depending on relative positions of R and B in the color wheel, the relative brightness (i.e., the grey scale) of R and B, and relative positions the of R and B on the “temperature” scale (e.g., a moderate warm color vs. a very cold color), the above 4 scenarios may leave audience with different emotional impressions, e.g., confliction, harmony, clear, confusion, lively, dull, …… Precise execution of a master color-scheme incorporating quantitatively correct combinations of coherent deployments (vectorial orders) of the scenarios a – d, across hundreds or even thousands of color pairs in a painting, holds the key to the success of a painting.

Planning of such a precise master color-scheme, precise execution of the master plan, and remembering everything in between have proven a set of tasks too heavy for human brains which are very much analog nature.

The new vocabularies that AI brought into the color language include: color gradient field, 2D array of vectorial color gradient, differential gradient field, and 2nd order vectorial color gradient.

It is not surprising, that Seurat often needed to work out multiples of studies before he released a final painting. For example, it took him 2 years and c.a. 60 studies to complete his famous “A Sunday Afternoon on the Island of La Grande Jatte”.

On the contrary, such a task is very suitable for AI which is digital in nature, with practically infinite memory capacities.

On a similar token, we can revisit the doctrines of the pointillism listed in paragraph 6 of this article, in the context of capabilities offered by AI. For instance, the pointillist theory points out that the emotion of joy can be created by predominance of luminous hues (i.e., the warm colors), which is correct. However, if we stop there, we would have missed a big opportunity unintentionally. AI points out this big opportunity to us.

Figures 3 and 4 compare the iconic pointillist painting “A Sunday Afternoon on the Island of La Grande Jatte” by G. Seurat (Source of Figure 3: https://en.wikipedia.org/wiki/A_Sunday_Afternoon_on_the_Island_of_La_Grande_Jatte) and an AI painting “API0048 《Color of Dusk in Barcelona · an AI’s View》” by ESAG (Source of Figure 4: https://www.esandag-ai-art-studio.com).

Figure 3 Figure 4

For a long time, the “A Sunday Afternoon on the Island of La Grande Jatte” has served as a textbook example to illustrate the pointillist theory, where a large %-area of the canvas covered with warm colors (e.g., greenish-yellow and pink in Figure 3), and sharp contrast (e.g., the sharp lines dividing the sun-shined lawn and shadowed areas in Figure 3) were deployed to make the painting look bright and joyful.

Granted, Figure 3 does deliver a bright and joyful scene, especially in the context of visual art in the 19th – 20th centuries. However, in the context of an AI age of the 21st century, if Figure 3 is compared with Figure 4, we can’t help but to notice that what Figure 3 delivered is not at an apex of what a painting can deliver. Using similar or even smaller %-area of warm colors (e.g., yellow, greenish yellow, and red), and less sharp contrast, but empowered by properly designed color gradients, Figure 4 has managed to create a higher level of the joyful emotion, although not necessarily at the ultimate apex.

Conclusion

(1) The color language used in the pointillistic paintings can be enriched and expanded, as empowered by AI, to include new kernels such as color field (i.e., 2D arrays of vectorial color gradients) and differential gradient field (i.e., 2D arrays of the 2nd order vectorial color gradients), in order to enhance the emotional impact and aesthetic values of visual arts (e.g., paintings), among other purposes.

(2) AI (digital and precise in nature) and human brains (analog and qualitative in nature) can augment each other. AI can help to clear the thick fog of mathematical computations to enable humans to see the possibilities that humans have failed to see in the past thousands of years. For a human artist living in the dawn of an AI era, energy should no longer be wasted on trying to compete with AI or channeled into a denial mode(s). The correct question that we should continue to ask is: how can we best embrace AI and use it to augment us in our creative endeavors.

(Written by Eric S. Shi, @ www.esandag-ai-art-studio.com)

作者简介

施舒，ES&AG 智能艺术工作室暨宜善达阁（在线）画廊的创始人。曾任职于一家大型跨国公司。2021年初离开公司，创办了ES&AG 智能艺术工作室暨宜善达阁（在线）画廊。在该工作室暨画廊里，目前除施舒之外，还有他制作的二位虚拟智能机器人助手协助他进行创作和管理。这二位助手是：ESAG和ESNA。ESAG的精力主要集中在创作智能绘画上；ESNA的精力主要集中在智能写作和画廊的智能管理上。

与普通以手作画的画家不同，施舒是目前为数不多的借助人工智能作画的画家之一。与一般借助人工智能作画的画家不同，施舒使用的人工智能都是由他亲手编程创作的。由于这一特点，施舒对于智能作画（乃至更广义的“知识创造”）有着一些与众不同的、更深层的观察和理解。

施舒曾留学美国（获化学博士学位，生物化学及生物物理学博士学位）和英国（获MBA学位）。