本文转载自微信公众号：曹雨西工作室

<人工智能山水图> 是我工作室新媒体艺术项目的新系列，这个系列主要目的就是面对当下利用科技流行词（Buzzword）对新媒体艺术作品进行包装鱼龙混杂的市场环境下，真正利用人工智能进行有力的视觉创作，进而呈现真实艺术创作和科学技术之间的关系，让不明觉厉新媒体艺术变得通俗易懂。

<Shan-shui Paintings by AI> is a new series from my studio’s recent new media art projects , this series are focus on how to make a impressive visual creation, present a realistic relationship between the technology and creative art and also shifting the ‘Abstractive Unknowingly Cool’ from new media art to more easy to understand concept  within an overwhelmly decorated by Tech-Buzzword art market.

<人工智能山水图>这个系列的技术核心是使用人工智能风格深度学习网络（SytleGAN）学习收集上至两千多张水墨图的像素数据后，利用学习训练后的模型进行自动无限创作模仿山水图样的创作。最后用二维码半吊图案（halftone pattern) 算法转化成可以被扫描的二维码图案。等于创造了一个可以无限制的创作水墨画风的二维码山水动画的机器人????️。

<Shan-shui Paintings by AI>  this series’ technical core is using SytleGAN to machine learning over 2000 collective Traditional Eastern Ink painting data, and create a model that can automatically generate Shanshui(Mountains and rivers) Pantings infinitely. And then converted the generated panting content into a scannable QR code pattern with halftone pattern technique. Which in conclude, to create a bot????️ that can using QR code to paints ink style Shanshui artwork infinitely.

在互联网收集到的人类手绘水墨画图库

Collection data of human ink drawings from internet

因为训练一个会模拟风格画图的机器人至少需要至少500到4000张风格参照数据图，所以寻找足够多的采样数据是做这个作品的第一个关键，不过感谢互联网时代#tag标签的流行，我可以快速地利用#水墨画#山水图等等迅速在全网寻找各种各样的图片材料，这其中小到一个初学山水画的小朋友习作照片，大到几百年前古代大师神作扫描数据，可以说是一个全人类关于水墨画意向的一个数据结晶。最后通过我人眼筛辨之后，大约找到了2000张左右的水墨画数据，其中大量的主要是以山水、花草大场景为主的水墨内容。

Because training a bot that simulates style drawing requires at least 500 to 4000 style reference data maps, So finding enough samples is the first key to making this work, thanks to the popularity of the #tag in the Internet era, I can quickly use # 水墨画 # landscapes and so on to quickly find a variety of picture samples on the internet. This range of the samples can start at as amateur as a beginner ’s landscape doodle in photo , or as professional as scanning document of maters’ artwork from hundreds of years ago. Finally, after being handpicked by myself, I collected about 2,000 pieces of ink painting data, which mainly consisted of landscapes, flowers and grass scenes.

互联网收集到的人类手绘水墨画图库

Collection data of human ink drawings from internet

后面的任务实际就是将把这些数据（图片们）交给AI StyleGAN（风格深度学习网络）进行训练和学习，这个过程基本就是等待，我大概等了四个小时后，学习后的模型就基本成形了。当第一次打开学习后的模组绘画的样板后，我还是被眼前的阵列图震住了：

The next task is actually to hand over these data (pictures) to AI StyleGAN for training and learning. This process is basically waiting. After waiting for around four hours, the learned model will basically take shape then. When I first opened the model drawing template, I was pretty shocked by the array of images generated by model in front of me:

人工智能根据训练后的模型自动生成的图库/Images array generatated by trained AI model 

正如上图所示，阵列里的每一张图都是人工智能根据训练后的模型自动生成而出的，反正我在看的时候发现每张图都可以看到不同风格、画派、时代以及技法的影子，甚至有时候机器人会模仿落款在画面的合适的位置写一些“火星文书法”。因为模型具有XYZ轴的阵列生成方向，所有每一次自动创作都会沿着XY轴做阵列生成。但是沿着Z轴的变化才是我的需求，虽然Z轴是可以理解成时间无法像XY轴一样直接立马显现出来，但是我可以以序列帧的形式渲染出图最后合成动画，而最后的合成效果着实令我震惊：

As shown in the picture above, each image in the array is automatically generated by the AI based on the trained model. When I looked at it, I found that each picture can see different styles, times, and the technique from traditional ink paintings world, even sometimes the robot will imitate the title and write some "Mars words calligraphy" in the appropriate position on the canvas. Because the model has an array generation direction of XYZ axis, all automatic creation will generate array along the XY axis. But the change along the Z axis is my demand. Although the Z axis can be understood as time but cannot be directly displayed like the XY axis, I can render the picture in the form of sequence frames and finally synthesize the animation, and the final composition really shocked me:

人工智能根据训练后的模型自动生成的动画

Animation clip generatated by trained AI model 

最终沿着Z轴变化生成的序列帧组成动画后，生成的山水图们无缝间地自动延展变化，每个像素仿佛生物变异一样和周围的像素们有序地进行舞动，最后视觉呈现的内容看起来也像几万年来山川河流地貌的变化，我不禁感慨人工智能模拟人类的真TMD可怕。

Finally, after the sequence frames generated along the Z axis change form an animation, the generated landscape images seamlessly automatically extend and change. Each pixel dances in an orderly manner with the surrounding pixels as if it is a biological mutation. The final visual content like mountains and rivers morphing remind me the plate tectonics the past millions of thousands of years on earth. I can't help feeling the potiential f**king power from AI.

将生成画面转化成二维码图案

Convertion from Image to QR pattern

后面部分的工作是将生成的“山水图动画”转移成二维码的图形界面，同时保留了二维码扫描的功能，最后将两个图层合并有机地结合“装裱”起来最终实现了项目的基本架构雏形。目前装裱之后准备了两个方向的版本：

The later part of the work is to transfer the generated "landscape image animation" into the graphical interface of the QR code, while retaining the function of QR code scanning. Finally, the two layers are merged and organically combined to "frame" to finally realize the project The basic structure of the prototype. After mounting, two versions have been prepared:

<人工智能山水图>_2x2阵列|Array_#A1时长|Duration：8'18" 
显示器阵列画框版本|Matrix Displays  Frame Version

一个版本可以算是传统架上的版本，这个版本目前的装裱设计是“土豪金”铝合金边框和四个方形显示器组成。这个版本是适用于个人或商务收藏，计划每套视觉限量一个，视觉时长8分18秒（对，非常吉利）。

One version can be regarded as the version on the traditional framing. The current mounting design of this version is composed of a "Rich Gold" aluminum alloy frame and four square displays. This version is suitable for personal or business collections. It is planned that each set of vision is limited to one, and the visual duration is 8 minutes and 18 seconds (yes, very auspicious).

‍

<人工智能山水图>_3x3阵列|Array_#B1

尺寸可变|Variable size

无限镜面沉浸装置版本|Infinite Mirror Immersive Version

另一个版本是无限镜面的沉浸装置版本（没错，打卡拍照好看），这个版本由LED面板和四面的镜面反射材料组成，尺寸可变支持定制。沉浸版本通过无限镜面的设置，将人工智能无限生成内容无限可能性和强大的学习能力通过镜面的无限反射寓意“侧面“地向观众表达出来。

The other version is the immersive version with the infinite refelection (yes, perfect for instagram photo feed). This version is composed with an LED panel and four-sided  reflective materials. It is with a variable size supports customization. The immersive version, through the setting of unlimited mirrors, expresses the power of unlimited possibilities of AI and  learning capabilities to the audience on "side "literally .

关于<人工智能山水图>的这个系列其他方向的版本正在计划和制作当中，其中第一个部分就是会给沉浸装置版本的二维码扫描之后的手机显示的互动内容有一个延伸的探讨。我希望观众拿起手机的时候不仅仅只是拍照打卡，可以通过这样的一个扫描行为有一些更深入的互动探讨

Other versions of this series of <Shan-shui Paintings by AIe> are in the plan and being produced. The first part is an extended discussion of the interactive content displayed on the mobile phone after the scaning of the QR code . I hope that when the audience picks up on their phones, they will not only take pictures and post on social media, but also do some more in-depth interactive discussions through such a scanning behavior.

和我另外一个作品<富春山居图>一样，二维码可以被扫描

Just like another work<Dwelling in the Fuchun Mountains>,QR code can be scaned

另一个我正在开发合作的版本是AudioVisual声音影像的现场表演版本，在这里也算是一个预告：目前我正在和古琴演奏家，跨界音乐人唐彬老师商讨合作完成一个表演版本的共同作品。下面这个视频是我将唐彬老师2017年的专辑《山水》的音乐片段和沉浸版本合成在一起，让各位先感受一下作品方向的意境和可能性，也算是为这次合作作品展示的一个小预告DEMO：

Another version I am developing and collaborating with is the live performance version of Audio Visual one. For now, I am discussing with Guqin player and cross-border musician Tang Bin to complete a joint version of a performance work. Here is i combine the music  from Tang Bin's 2017 album "Landscape" with the immersive version to intorduce the mood and possibilites of the direction of the work. Thie video below can also be regarded as a small preview DEMO for this collaborative work:

声音影像表演版本预告/Audio Visual Performance Version Trailer

音乐：唐彬 / Music: Tang Bin

<人工智能山水图>关于这个作品系列更多的未来消息和更新内容，我会继续更新在公众账号的新帖子中，欢迎关注我的公众账号，如果你有更多的问题欢迎留言这个帖子我会及时回复。最后还要声明，本文章内容和文章内作品<人工智能山水图>已经注册著作权和版权保护，受中华人民共和国著作权法、中国版权中心和美利坚合众国著作权法保护，未经曹雨西工作室允许禁止转载。

For more future news and updated content about <Shan-shui Paintings by AI> , I will update the new post in the public account, so please fellow my public account, if you have any questions, please leave a comment below and I will reply in time. Finally, this article and the artwork <Shan-shui Paintings by AI> in this article have been registered with copyright and copyright protection services, protected by the Copyright Law of the People's Republic of China, the Copyright Center of China and the Copyright Law of the United States of America, and are not allowed to be reproduced or reposted without permission from Cao Yuxi Studio.

‍

署名－非商业使用-禁止演绎 (by-nc-nd)

2014-2020 曹雨西工作室 © 版权所有

2014-2020 CAOYUXISTUDIO © ALL RIGHT RESERVED