Telling a story, narrative, in a painting is one of its most common purposes, and greatest challenges. A landscape painting shows a view at a moment in time, but doesn’t normally tell a story, as that requires a minimum of two states, with the story linking them. So, although we might speculate what’s going in that countryside, without an indication of what went before, or what happened afterwards, there’s no story there.

Over the centuries, even millennia, since humans have been painting, several techniques have developed for telling a story with more than one timepoint shown in paintings. Although the terms used have varied, in general those fall into the following categories:

• instantaneous, where the image is intended to show what was happening at a single moment in time, even though it’s likely to contain references to other moments in time;
• multi-image, where a series of separate images (paintings) is used to tell the story;
• multiplex, where a single image contains representations of two or more moments in time from a story;
• multi-frame, where two or more picture frames are used to tell a story, as in comics or manga;
• polymythic, where a single image contains two or more distinct stories.

In this article and tomorrow’s sequel, I show examples of each of these.

Instantaneous

Nicolas Poussin’s Rinaldo and Armida from about 1630 draws its narrative from one of his favourite literary works, a then-popular epic poem by Torquato Tasso (1544-1595) titled Gerusalemme Liberata (Jerusalem Delivered), and published in 1581. I have written a series of fourteen articles showing paintings and telling its story, which start here. This particular episode is detailed here.

The sleeping knight is Rinaldo, the greatest of the Christian knights engaged in Tasso’s romanticised and largely fictional account of the First Crusade, who has stopped to rest near the ‘ford of the Orontes’. On hearing a woman singing, he goes to the river, where he catches sight of Armida swimming naked.

Armida, though, had an evil aim, in that she had been secretly following Rinaldo, intending to murder him with her dagger. As the ‘Saracen’ witch who is trying to destroy the crusaders’ campaign, she had singled out its greatest knight for this fate. Having revealed herself to him, she sings and lulls him into an enchanted sleep so that she can thrust her dagger home.

Just as she is about to do this, she falls in love with him instead, and that’s the instant, the twist or peripeteia (to use Aristotle’s term), shown here. A winged amorino, lacking the bow and arrows of a true Cupid, restrains her right arm bearing her weapon. Her facial expression and left hand reveal her new intent to enchant and abduct him in her chariot, so he can become infatuated with her, and forget the Crusade altogether.

This is a single moment in time, in which Poussin has ingeniously incorporated references to the past and future. Provided that you’re familiar with Tasso’s story, it’s a superb example of instantaneous narrative, as practised throughout the history of painting across all continents and cultures.

Jean-Léon Gérôme’s Cleopatra before Caesar (1866) also depicts a single instant, but again has references to prior events, particularly the screwed up carpet, used by Cleopatra to gain entry. Her dreamy look towards Caesar also anticipates her affair with him. It therefore has instantaneous narrative.

Sometimes paintings with instantaneous narrative can make quite small and subtle references to other events in the story, and confirm their narrative nature. In Edward Burne-Jones’s Cinderella (1863) the only such reference is the missing slipper on Cinderella’s right foot.

Although Édouard Detaille’s Le Rêve (The Dream) (1888) contains two images, these aren’t in fact linked by normal narrative, but the dream image shown in the clouds could be considered as a form of analepsis, or flashback, making it instantaneous narrative.

Multi-image

I’ll be brief with these, as I have covered more examples here and here.

In 1856, Arthur Hughes told the story of The Eve of St Agnes in this triptych, read from left to right. At the left Porphyro is approaching the castle. In the centre he has woken Madeline, who hasn’t yet taken him into her bed. At the right the couple make their escape over drunken revellers.

Akseli Gallen-Kallela’s triptych showing the Aino Myth (1891) contains three separate images telling one of the stories from the Kalevala myths. It is therefore multi-image narrative, within which each image is itself conventional instantaneous narrative.

Multi-frame

Multi-frame paintings are by no means uncommon, but most usually adopt rectangular or square form. Indeed many of the more spectacular frescoes are in effect multi-framed, where there are several images on a single continuous surface. This is similar to the more recent development of comics/BD/graphic novels.

Gaudenzio Ferrari’s Stories of The Life and Passion of Christ (1513) arranges twenty frames covering the life of Christ around a central frame with four times the area of the others, showing the Crucifixion. The frames are naturally (for the European) read from left to right, along the rows from top to bottom, although the Crucifixion is part of the bottom row. This is a layout which is commonly used throughout graphic novels too, of course, and is a superb example of multi-frame narrative more than three centuries before Rodolphe Töpffer started experimenting with comic form.

The four separate episodes forming Hans Sebald Beham’s Scenes from the Life of David (1534) are arranged in a square, so that each occupies a triangular frame, clearly separated from the others, and quite different from a normal linear layout. The snag with this is that the panel is really only suitable for viewing when laid flat on a table, otherwise only one of the frames is correctly orientated. Beham clearly liked the symmetry afforded by this layout, and enhanced it in his composition of the two frames shown here at the top and bottom.

Frans Francken the Younger’s The Crucifixion of Christ, with Scenes from the Life of Jesus (1600s) puts the Crucifixion scene at the centre of a rectangle, around which are twelve scenes from the life painted in either normal or brown grisaille. Unfortunately those peripheral scenes are difficult to differentiate from one another, thus to identify, but they appear to be read in a clockwise direction from the upper right, rather than linearly.

Polymythic

Although linked, and often told together, the stories of Echo and of Narcissus can be separated, and it’s therefore feasible to classify John William Waterhouse’s Echo and Narcissus (1903) as being unusual in showing polymythic narrative.

A few paintings appear even more complex: Velázquez’s Las Hilanderas (The Spinners) may contain one narrative in the foreground, a second in the background, and a third in the painting of The Rape of Europa shown in the far background. This would make it polymythic narrative at the very least.

在上个月去上海之前，@取景框看世界 在微信上邀请我一起做一期关于设计行业的科普视频，面向学生群体做一次对行业整体状况的分享。这次是 B站 向他发起的约稿，也是他频道的主要内容类型之一。答应他后，从上海回来我就根据自己这么些年的体会和反思，录了一期比较掏心窝的内容。由于参与的人比较多，直到前天，节目才终于上线。

因为参与的人数比较多，我说的话也比较多，所以在汇总的成片里需要剪掉一些。但我又觉得难得录一期视频（从去年11月到现在都没有更过视频了），为了保证表达的完整性，我还是得有一个自己的版本。但因为实在忙得顾不过来，于是麻烦 @小雨 帮我把这条视频剪了出来。

毕竟是 @怪物尚志 的御用摄影/后期，有他的帮忙，我的视频里头一回多了许多配合文案的动画，比我一个人在镜头前单口相声好看多了。这一期花絮，也在昨天发了出来，跟正片错开一天。

这期视频的封面是我昨晚用 midjourney 画的。

在这条视频里，除了科普工业设计的一些基本概念和行业现状，我也特别聊了一下 AIGC 和设计师之间的关系。前两个月虽然一直很忙，但我也一直在关注 AIGC 的动向。这两个月的变化实在是过于惊人了！

以下是我去年八九月用 midjourney 画的一些东西：

然而在这短短半年内，版本已经从当时的 v3 发展到了现在的 v5。ChatGPT 也从 GPT-3 发展到了 GPT-4，坊间传言 GPT-5 的研发已经完成且爬完了全网所有视频，可以直指某一条视频中的某一段话，只是还没发布。可以说开年后的这两个月内，AIGC 的发展速度已经大大超乎了预料，甚至正处于失控的边缘，它们正在开始颠覆一些东西。将来会发生什么，无法预料，但一定有什么事情已经在发生了。

所以我觉得，无论如何，再忙也得重新用起来。光是跟进各种新闻和消息是不行的，midjourney 前两天刚发布了由图片转译成 prompt 的新功能，多模态的 AI 已经不远了，这会更进一步推动人和 AI 之间的交互。我觉得，自己还是得保持使用状态才行。所以前两天我又重新充了值，开始体验它的新版本。

我先试着画了一些机甲的东西，例如这样的：

上面的两张的用 prompt 直接生成的，但下面的两张，是用 /remix 命令修改了部分描述词后的新图。可以看到下面的图和上面的图保持了相关性，于是我想试试，如果用这个命令替换背景会是什么效果，于是有了这组车的图：

上面的两张图是用 prompt 直接生成的，当我用 /remix 替换了背景描述的 prompt 之后，就生成了下面的两张。更换环境之后，车辆的姿态和镜头视角几乎没有变化，车身的反光与环境之间的关系也很自然，这个效果已经可以说非常惊人了！

然后，我随便画了一些白色的机器人站在燃烧废墟上的场景：

用 /remix 替换了机器人配色部分的描述词后，生成了下面这样的图：

对 /remix 有了基本的体感之后，我开始尝试用 /blend 命令来做一些融合的实验。

首先，我随便描述了一个赛博少女，得到一些随机的图：

接着，我再随便生成一个红发少女：

材料准备好之后，开始把它们进行组合。

第一次先尝试融合两张图，一个是游戏画风的机甲人，一个是二次元的赛博少女。

它们俩合成后，得到了以下这个人物：

新角色具备其中一张图里人物的长相特征与体态，也有另一张图的配色和机甲特点。虽然得到的结果具有随机性，但既然可以这样融合，那么应该也可以通过 /blend 命令来得到一些更有目的性的创作。

有了第一次的体验后，第二次我用三张图片进行合成：

图一是现画的半透明金属机器人，图二是上面准备好的红发少女，图三是现画的骑士。

这三个合成出来的新角色，同时具备了细碎的金色细节、波浪红发、银白色盔甲：

但这不是我想要的，我想试试加大红发少女的比例。在垫图的方式下，可以通过 –iw 命令来分配各个图片之间的权重占比，但是在 /blend 中不能这么操作。于是，我想通过把合成的新图作为素材，再一次与红发少女进行融合，并加入机甲的元素来强化她身上盔甲的质感。

二次合成使用的图，如下：

合成出来的新角色我非常满意！

她既有红发少女面部和眼神的特征，又把两副银白外甲融合得非常优雅，也保留了初始半透明金属机器人遗传下来的金色金属关节的特征，又做出了图三机甲的坚硬感和图一外甲的银白光泽。这一次的融合很成功。

但如果 /remix 可以局部替换特征，那么这些没有写 prompt 而是通过 /blend 直接合成的图，能否通过 /remix 加入新的 prompt 来修改已有的特征呢？

为了让实验效果明显一些，我想让盔甲的白色部分比例缩小，增加金属部分的比例，于是就先把这批图重新刷了几遍，直到出现肩甲是金色的变异版本：

然后不断在此基础上进一步变异，强化金色肩甲的特征：

准备好之后，我在 /remix 中添加新的 prompt：pink armor

以下是修改特征后的结果：

整体的效果我还是挺满意的。一来，新生成的人物很好地保持了最初红发少女的眼神和神态；二来，金色金属被替换成粉色金属后，金属质感的表达是正确的。虽然头发也一起变成了粉色，这确实是没完全理解指令，但原有的发色搭配新的粉色盔甲也确实不是很和谐。

到此，重新开始用 AIGC 工具的热身完毕，找回一些感觉了。

至于这期视频封面里用到的车图，是我昨晚用 Maserati 和 Ferrari 以及 Apple 和 Tesla 分别杂糅出来的缝合怪。虽然乍一看好像没什么新奇的，但是如果我把去年八月底用 midjourney 画的汽车拿出来对比，就会意识到这是多么疯狂的进化速度了：

上面三个是去年八月用 v3 画的车；

下面这些是昨晚用 v5 画的车：

虽然工业设计有大量的细化和落地工作是 AI 无法干的，但从目前来看，无论是 midjourney 还是 Stable Diffusion + controlNET 都已经可以很好地帮助设计师完成概念发散和快速枚举了。这样的图像质量，通过垫图、remix 和 blend 的组合使用，完全可以在创意初期快速拉出一批高质量的「草图」，设计师可以把更多的精力放在对方案思路的推敲、对细节的考据以及各个环节的沟通协调上。

从今年二三月开始到往后的十一二年，人类社会将迎来一场以破坏为开端的变革和创新。

无论我们是否愿意，都将一起进入新的世界。