Midjourney for Architects

Today I would like to tell you how I use Midjourney if I want to generate a good image as a reference for an architectural project. I must say that besides this there are several other uses such as creating textures, icons, backgrounds for renders etc., but this is another topic. In general Midjourney is configured to make more or less pretty pictures almost always even if you put just a single word to the prompt. But although Midjourney will give you pretty good images you will quickly realize that the results seem to be similar, if you ask about specific architectural styles or use the names of star architects in your prompts. My advice: don't start with architecture. The next method is often used by artists even in non AI workflow: we will start with the references from nature: a tree leaf, starfish, snow, tornado, steam movement... First try to generate an image on a topic that inspired your project, but not directly related to architecture. For today's examples I will take the human brain or rather brain neural network. 

Let's start with this prompt:

human's brain neural network topology visualization on the black background --q 2 --v 4 --upbeta

it's not bad, but it's not what I need as a reference, I need more abstraction, so we'll continue.

To get the image that you have imagined can be tricky, you need to go through many words and their combinations, and there is no good advice here, you just have to guess, though of course the logic of guessing is not random. In the example that we are parsing, the word combination macro photo or macro view helped to create the proper structures. It seems logical, but it took me a lot of experiments to get it right. The full prompt looks like this:

human's neural network topology visualization on the black background, macro view --q 2 --v 4 --upbeta

We have obtained the desired reference. Here we must understand that the reference is not a particular image, but the fact that your prompt produces interesting images that can be a reference for further architecture production. That is, the actual reference is the prompt and all the possible results that can be generated by this prompt. Well, from this point we can start adding something architectural to the pot. let's try it this way:

[human's neural network topology visualization macro view]::3 + [3d volumetric steel structure]::1 --q 2 --v 4 --upbeta

First we put the reference and take it in brackets, then we add the second brackets that describe our architectural problem. After each bracket through a double colon we put weights, here 3 and 1, that is, the reference weight is three times more important than our architectural component, and the result gravitates toward abstraction as brain neural network rather than steel structure.

It is logical to further try to play with the weights, you can swap them or set other proportions:

[human's neural network topology visualization macro view]::1 + [3d volumetric steel structure]::3 --q 2 --v 4 --upbeta

We have already gone through several iterations, but we don't have any buildings in the pictures yet, that's how it's meant to be, don't worry)

Now let's add some more brackets to make it clear that we try to get not just a steel structure, but a building in a certain style and in a certain place. Again, we can play with weights to make the result more or less abstract. 

[human's neural network topology visualization macro view]::2 + [glass facade with 3d volumetric steel structure]::1 + [futuristic building in the forest]::3 voxels, unreal engine --q 2 --v 4 --upbeta

Proportions of the image have a very strong effect on the result, which can be set through the parameter --ar. In other words, if you want to get a skyscraper, not only add the appropriate word in the prompt, but also stretch the image vertically.

[human's neural network topology visualization macro view]::2 + [glass facade with 3d volumetric steel structure]::1 + [futuristic skyscraper]::3 voxels, unreal engine --ar 2:3 --q 2 --v 4 --upbeta

The rest is a problem of customization, and although it can take hours, there are no special tricks here, just experiment with the weather, colors, entourage, image style... For example, I want to add the downtown atmosphere, snow, some bright colors, so write it all in the prompt. And don't forget about the Remix feature, which is available from the Midjourney settings. You can start with weights, like in the example below, and then change them with the Remix variations.

[colorful human's neural network topology visualization macro view]::3 + [3d volumetric steel structure]::2 + [futuristic building in the city]::3 + [winter day, snow, downtown]::2 + [8k, fujifilm photo style]::2 --q 2 --upbeta

Of course you can use an image reference in the prompt instead of a text reference, but sometimes you don't have the suitable image, and it's easier to generate it first. And the Remix function works on the basis of the previous result, i.e. it is more or less like using the image directly in the prompt. 

This is the method. I hope it was useful and interesting)