top of page

Course in Collaborative Image Generation

Ai Drawing Tools as Typological Libraries

First of all, I’m biased. I’ve been doing landscape drawings in a semi-professional context since 2008 and professionally since 2014.  Some of my drawings have even become built parks! I’ve also taught landscape representation at Harvard GSD Project Link, the Center For Court Innovation in NYC, and in Myoshinji zen monastery in Kyoto.  The experience teaching watercolor as a rendering tool in zen gardens was particularly formative - we had to consider a constantly flickering feedback loop between perception, illusion, and shared reality. All this raises the question today: can Ai visualization platforms actually improve our work as artists and designers? Contrary to the immediate assumption, Ai drawing tools may actually embolden hand sketching - real-time drawing - as a tool of design.



"Ai drawing tools may actually embolden hand-sketching - real-time drawing - as a tool of design."

Alan Waxman 2/23/2023

Ai presents typological libraries - into which you can feed your drawings.  These typological libraries are a set of public fantasies which may, if used enough, change how you see the world.  Working back and forth between your own experience and the hive-mind of the library, actual drawing abilities and fluidity of creativity grow in importance.  The risk, is that your own work may become entwined in public ways of seeing. There is a common perception that Ai produced drawings are novel- this article, and various court cases - may show this is a bit of a stretch.  But Ai still performs a valuable service, assembling typological representations for you to view and work with.

What is a typological library?

A typology is a “form.” In the most platonic sense each form is “essential” and it can only be seen as a representation, a kind of shadow of the “real” thing.  Even a written word or spoken word is only a representation of the true idea, or so we are led to believe by Plato in the Timaeus and the Allegory of the Cave. 


Once we understand that we never can actually see a form, only it’s myriad representations, then we can begin to understand how a typological library assembles a variety of representations according to a particular topic.  Consider the metaphor of a physical library.  You find a section labeled “Dogen Zenji” and below the heading are potentially hundreds of books on Dogen and by Dogen. In these books there are many representations of Dogen and the ideas of Dogen.  


Likewise, in a library if you search for “melancholy” you will find many books, representations, and pictures of melancholy.  You might find some great anthologies such as “The Anatomy of Melancholy” by Robert Burton, but there will never be a singular definitive answer. No library is complete.   This set of data, in the library, can be referred to as a “training set.” It presents a vocabulary, a veritable thesaurus of representations of “melancholy.” 

Midjourney Ai and Google Image Search

Midjourney Ai represents a particular typological library.  Google Image Search also represents a typological library.  In no way do either platforms represent “reality” fully.  Midjourney Ai is biased towards art from the 17th to the 20th centuries primarily in the Western tradition. There is a strong emphasis on biophillic imagery a la Ruskin, certain pre-Raphaelite elements, and a strong influence from impressionism.  In a sense, Midjourney Ai is like a singular pop artist today. One who uses many famous precedents, and never cites their work. However, the only notable point is that the artist is not you, because it’s not a real time drawing process.  However, you are a good collaborator, or design director for the artist.  Midjourney provides a set of typological elements based on key words which you as the design director enter.


Google Image Search is very similar to Midjourney Ai and other Ai drawing tools.  However, Google Image Search provides you with a variety of images when you enter a prompt, rather than one image.  Google Image Search also always provides a citation of the publication of the work in the form of a hyperlink, which may or may not suggest the original artist.  Google Image Search and Midjourney Ai both provide very valuable typological libraries.

"Once we understand that we can never actually see a form, only it's myriad representations, then we can begin to understand how a typological library assembles a variety of representations according to a particular topic."


Part of the "value" of Midjourney Ai is that it is abstracted and citation is removed.  This postmodern dada-esque collage technic allows a kind of conceit in which the typological representation is potentially believable as being close to the “essence” of the idea.  I repeat, this is a conceit.  As an artist, I would urge you to take the typological representations presented as precedents to actually enter the real time art making process yourself, drawing, or making a collage of these images in Photoshop. 


One might argue that this is unnecessary, because Midjourney, and the like, already does the collaging, and that your involvement in the process as a craft is unnecessary? My response is that there is a difference between going through a library or book and finding the image you like based on text, and then saying it’s yours; and actually drawing or collaging yourself.  The difference can only be felt - because real time drawing is something you feel while doing.

"This postmodern dada-esque collage technic allows a kind of conceit in which the typological representation is potentially believable as being close to the 'essence' of the idea."

Design with tools (Ai)

As a designer, however, there is another consideration, beyond drawing - that is how the image is received towards the goal of the design, and how feedback changes your deployment of the design. 


Consider an ecosocial design project in which you are responsible for helping a community to reduce diabetes in a neighborhood.  Any drawing tools you use are deployed in order to help the community visualize some of the issues they face.  If they can visualize them and map them out, then they can better personally and collectively address them. 


In meetings, participants may use words to describe how they feel about a space: 






seen better days, etc, 





up and coming etc.  


These terms may be specifically applied to a kind of space: 

the area behind my house, 

the lot over there, 

the abandoned junk yard, 

the KFC, 

the day care center, 

the mountains, 

the flooded area by the rail yard, 

the largely unused office buildings, etc. 


More terms might be used:

I can’t get where I’m going 

I’m stuck 

It’s just a loop 

The system is rigged 

Be yourself 

Find internal satisfaction 

Teaching and sharing knowledge and joy 


As a designer it would make sense to make your own drawings as specific as possible, especially when visualizing an intentional outcome in a community.  However, in the process of helping a group of participants through the largely psychosocial process of visualizing and articulating how they feel about a space, and how they might feel about things that happen there, using typological libraries can be very helpful. 


Here are some combinations from Midjourney Ai and from Google Image Search produced by a few of the combined prompts provided by the hypothetical participant group above.  Note the Google Image Search images are literally the first to come up on search unless otherwise noted.


Abandoned area by my house.png
Sad flooded area by the railyard.png
Screen Shot 2023-02-26 at 9.15.48 AM.png


Google Image Search: “The system is rigged”

Stuck in the abandoned lot over there.png
forlorn dirty kfc.png

Midjourney: “Forlorn dirty kfc” 



Midjourney Ai: “Abandoned area behind my house”



Midjourney Ai: “Sad flooded area by the railyard”



Midjourney Ai: “The system is rigged”



Midjourney Ai: “Stuck in the abandoned lot over there”

Day care center internal satisfaction.png


Google Image Search: “Day care center internal satisfaction”



Midjourney Ai: “Day care center internal satisfaction”

Be Yourself Blooming Mountains.png


Google Image Search: “Be yourself blooming mountains”



Midjourney Ai: “Be yourself blooming mountains”

Forlorn dirty sharing knowledge.png


Google Image Search: “Forlorn dirty sharing knowledge”

Abandoned area sharing joy.png


Google Image Search: “Abandoned area sharing joy”



Midjourney Ai: “Forlorn dirty sharing knowledge”



Midjourney Ai: “Abandoned area sharing joy”

These images can all be “refined” through the process of bringing them back into the platform and the choosing more subsequent similar images that result. 


From the standpoint of design process, all these images may be valuable to help work through a shared psychosocial process, or more likely, some of the images are more interesting to various participants than others.  The important thing is to keep working through the process of drawing out the humors or the participant group, such that the group can identify trends that the collective wants to work through in a design project.  Then the collective can begin to articulate what are the specifics of a proposed project, and the process can be repeated, working more and more towards the proposed project.

Reinterpreting sketches


Now, let’s say that you are doing a design process and participants, such as yourself, have described the scene that they want to create.  You can use Ai drawing tools to add psychosocial conditions desired to sketches you have already made.  Let’s try looking at how Midjourney Ai and Google Image Search reinterprets sketches.  I’m going to use two slightly different sketches, Mountain 1 and Mountain 2.


Sketch of a landscape, hand drawn by me, contrast and paper removed by photoshop and illustrator



Mountain 1

Reverse image search mountain 1.png


Mountain 2

Reverse Image Search Mountain 2.png

The typological libraries of Ai really just allow you to interpret these sketches through the different psychosocial atmospheres we already proposed.  Probably for legal reasons, Midjourney Ai doesn’t allow itself to simply reinterpret your single drawing freely, or share referenced drawings.  However, you can combine your drawing with prompts so that they act as “filters.”


Because of some issues with the software, we will have to define the drawing style so that it’s unified. In this case, the hand drawings are a kind of line drawing brush painting medium, so this will need to be added to script. We will also only feature one of the four options presented in Midjourney and beside it the original drawing and the “filter” prompt typologies. Now let’s see this landscape through the “filters” associated with Midjourney Ai already discussed above:



Mountain 1



1.Mountain 1 + Forlorn dirty kfc line drawing brush painting:



Mountain 2



2.Mountain 2+ forlorn dirty kfc line drawing brush painting:



Mountain 1



3.Mountain 1+ Abandoned area behind my house line drawing brush painting



Mountain 2



4.Mountain 2+ Abandoned area behind my house line drawing brush painting



Mountain 1



5.Mountain 1+ Sad flooded area by the railyard line drawing brush paintingline drawing brush painting



Mountain 2



6.Mountain 2+ Sad flooded area by the railyard line drawing brush painting line drawing brush painting



Mountain 1



7.Mountain 1+ The system is rigged line drawing brush painting



Mountain 2



8.Mountain 2+ The system is rigged line drawing brush painting


9.Just to have a little fun with a filter of a filter, consider

+ studio Ghibli) (upscale version two) + surreal color evening volumetric lighting Jakuchu (upscale version four)



10.+ 1920s travel magazine klimpt



Mountain 1



11.Mountain 1+ stuck in the abandoned lot over there line drawing brush painting



Mountain 2



12.Mountain 2+ stuck in the abandoned lot over there line drawing brush painting



13.+Frazetta ghibli early morning luminescence (upscale two)



14.+cinematic still shot [princess mononoke]



Mountain 1



15.Mountain 1+ Day care center internal satisfaction line drawing brush painting



Mountain 2



16.Mountain 2+   Day care center internal satisfaction line drawing brush painting



Mountain 1



17.Mountain 1+ Be yourself blooming mountains line drawing brush painting



Mountain 2



18.Mountain 2+  Be yourself blooming mountains line drawing brush painting



Mountain 1



19.Mountain 1+ Forlorn dirty sharing knowledge line drawing brush painting



Mountain 2



20.+frezetta studio ghibli bright night sky overhead cinematic still blade runner realistic 4k



21.Mountain 2+ Forlorn dirty sharing knowledge line drawing brush painting



Mountain 1



22.Mountain 1+ Abandoned area sharing joy line drawing brush painting



23.+in style of ralph bakshi



24.+wide angle shot fish eye lens hyper realistic video still



Mountain 2



25.Mountain 2+ Abandoned area sharing joy line drawing brush painting

Midjourney Ai effectively filters the original drawing or chosen imagery through a variety of typological humors, classified by various kinds of historical artwork, whether explicitly identified or not. By comparing the images generated entirely by the Midjourney typological library and also the Google Image Search typological library found in search with the adapted hand drawings, its possible to pretty clearly see how the images are produced by adding the typology to the drawing through the diffusion technique.

For example the daycare image      +         Mountain 1          =.          the filtered composite  


The "abandoned area behind my house" + Mountain 2          =.          the filtered composite  


The fact that these are expected results are facilitated by the feedback system in the program, intending to please the user with just enough blend and just enough slight changes.


Typology, for the purposes of design, much like the “essence” it is supposed to represent, is itself only a small part of design process.  One of the wonderful opportunities that working with typological libraries of Ai presents is that we can reverse image search what we find to further identify who our collaborators are across the hive mind. 


Reverse image search on several of the above images delivered a few similar pieces to several (7,11,13,20,24)  “created” in Midjourney above.

Screen Shot 2023-02-26 at 10.25.24 AM.png

How many of these images were also created with Ai? It’s impossible here to find out. But we can only really rest assured a drawing is crafted by an author when it is hand drawn or painted. Regardless of authorship, the selection of different atmospheres and humors or “filters” can be an important part of coming to be a better understanding of a set of psychosocial conditions, particularly if the typologies are selected by a collaborative group of participants.  It’s up to the designer and other participants to come to an understanding of the significance of the imagery in terms of the larger process of changing some desired outcome.

bottom of page