0% found this document useful (0 votes)
24 views87 pages

Promptbook Ebook

The document is a comprehensive guide titled 'The Promptbook', focusing on using AI for image generation, particularly through tools like Midjourney and Stable Diffusion. It covers various aspects such as the anatomy of a good prompt, available AI tools, and practical examples for generating images. The authors, Diana Turčeková, Lukáš Pelant, and Kacper Borysewicz, aim to provide readers with practical insights into the art of AI-driven image creation.

Uploaded by

moviepitchdeck
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
24 views87 pages

Promptbook Ebook

The document is a comprehensive guide titled 'The Promptbook', focusing on using AI for image generation, particularly through tools like Midjourney and Stable Diffusion. It covers various aspects such as the anatomy of a good prompt, available AI tools, and practical examples for generating images. The authors, Diana Turčeková, Lukáš Pelant, and Kacper Borysewicz, aim to provide readers with practical insights into the art of AI-driven image creation.

Uploaded by

moviepitchdeck
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 87

Diana Turčeková, Lukáš Pelant, Kacper Borysewicz

THE
PROMPTBOOK USING CHATGPT AS A VIRTUAL COLLABOLATOR
1
Using AI to generate perfect images
Course final project, 2023 - All Rights Reserved
1. INTRODUCTION ………………………………………...…………………3
1.1. Authors ……………………………………………...…………………....3
1.2. Available tools ……………………………………………………….…4
1.3. Introduction to prompting ………………………………………....11
1.4. Fantasy illustrations …………………………………………..…….14
2. RUN YOUR OWN AI …………………………..………………………17
2.1. What is it about? ………………………………………................…17
2.2. Demo version …………………………………………………………18
2.3. Download and installation ………………...………………………19
2.4. Image generation ……………………………………………………21
3. ANATOMY OF A GOOD PROMPT …………………………....22
4. SUBJECT ……………………………………………………………………23
4.1. Subject definition ………………………………………………….…23
4.2. Negative prompts ……………………………………………………31
4.3. Keyword strength ………………………………………………...….33
5. RESOLUTION ………………………………………………………….....37
6. LIGHTING & RENDERING …………………………………………49
7. STYLE ………………………………………………………………………...57
7.1. Chaos …………………………………………………………………....57
7.2. Seeds …………………………………………………………………….59
7.3. Stylize ……………………………………………………..…60
8. MEDIUM ……………………………………………………………………..65
8.1. Specify the medium ………………………………………………...65
8.2. Sketch …………………………………………………….…68
8.3. Combining mediums ………………………………………………..71
9. ARTIST ……………………………………………………………………….72
9.1. Artists style …………………………………………………………….72
9.2. Narrations ………………………………………………...…74
10. IMAGE PROMPTING ………………………………………………….76
11. BONUS ……………………………………………………………………….86

2
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
1. INTRODUCTION
1.1.Authors

Welcome to the fascinating world of image generation by artificial


intelligence! We would like to invite you on a journey through the
extraordinary possibilities offered by this innovative field of art. Our
book, entitled "The Promptbook” is the result of the collaboration
of three talented students from three different countries.

Diana Turčeková Lukáš Pelant, Kacper Borysewicz,


Slovakia Czech Republic Poland

Together, as a team of authors, we want to present readers with


practical aspects of image generation using artificial intelligence.
Enjoy a wonderful journey through the fascinating world of image
generation by artificial intelligence!

Diana, Lukáš and Kacper

3
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
1.2. Available tools

There already exists a vast amount of different AI tools for


generating images. We tested the most popular ones to figure out
which one is the best for different purposes.

When we began collecting data for this book during February 2023,
we found out that it is better to use different tools for different tasks.
For example DALLE-E excelled at paintings and animal
illustrations, Midjourney at photorealistic rendering, Nightcafé was
the best for artistic stylization and Jasper art for using different
medium styles.

But very quickly Midjourney started to dominate the market and


with its frequent actualizations and connection to Discord, it
became massively superior. We, all authors, agree that its
elaborate command system and the ability to understand what you
want from it is unmatched.

For this reason, we briefly present to you some of the AIs we tried,
but most of the book is based on experiments with Midjourney, as
after you understand the prompting fundamentals, whatever you
want to do - commercial images, illustrations, logos, photo editing
or designing your game - Midjourney is most likely going to be your
choice number one.

What can you try?

Midjourney is the way to go. Up to this day, it is clearly the most


powerful and precise image generator. It is accessible via Discord
application and although it is subscription based, it is very
affordable.

4
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
DALL-E 2 is from our perspective right behind Midjourney and you
can create a lot of realistic pictures or well stylized illustrations and
paintings. If you use, now very popular, Bing AI, it has its own
image generator, but it actually runs on DALL-E as well. It is
important to mention that it comes directly from OpenAI, so it is
very easy to connect it with ChatGPT.

5
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Dream by Wombo
is probably the best
image generator
that you can easily
use with your
phone, but it still
likes to add more
limbs now and
then.

CrAIyon is a good
experimenting
alternative. It is not
as precise as the
other ones and it
takes longer time
periods to
generate images,
but for now, it is
completely free.

6
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Stable Diffusion is almost comparable to Midjourney and it is used
by many different providers. It was developed by Stability.AI and it
is widely used. Later, we are going to teach you how to get
unlimited access to this one, so you can practice your prompting.
Different providers use different settings, so the results are still
going to differ. You can try Dream Studio, which is the official web
provided by Stability.AI. You can try different versions of Stable
Diffusion here.

https://beta.dreamstudio.ai/generate

Replicate is used for sharing and prompts and generated images


by Stable Diffusion V.2.

https://replicate.com/about

7
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
PlaygroundAI has a lot of tools to adjust the generation process.

https://playgroundai.com

As Stable Diffusion is open-source, you can now access it via


Google Colab thanks to some very enthusiastic users.

8
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
https://colab.research.google.com/github/qunash/stable-diffusion-
2-gui/blob/main/stable_diffusion_2_0.ipynb

BaseTen is a platform for startups and they recently released API


support for Stable Diffusion. There are no settings, but it free for
now.

https://app.baseten.co/apps/VBlnMVP/operator_views/nBrd8zP

9
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
And to still look at something other than Stable Diffusion, Creative
Fabrica uses a well done engine for imitation of different art styles.
It is free to use for small resolutions, but there is a small amount of
prompts it can understand at the same time.

https://www.creativefabrica.com/spark/

Fusion brain is another interesting free project. It uses a very


small amount of seeds, but the results are still beautiful. Only the
variety is limited and it seems like it is trained on smaller amounts
of data. It is awesome for logos.

https://fusionbrain.ai/diffusion

10
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
1.3. Introduction to prompting

To begin with, let's see an example of the evolution of a prompt from basic
to very elaborate and how important it is to word prompts correctly. Let’s try
to make some pictures for a Nike online shop using Dream studio & Stable
Diffusion

„A nike shoe“

“A nike shoe,
dramatic light”

“A nike shoe, studio light”

11
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“One shoe of Jordan 1 nike in Yayoi
Kusama style, side view in the
center, highly detailed”

„Nike pace shoes made of rubber and


composite fibres, black Nike logo in the
shoe, photography by Steven Meisel,
beautiful lighting, unreal engine, octane
render, hd, iq 180 --ar 2:3 --style 4a“

“Nike Air Flight Huarache Shoe


pride month, bright, colorful,
glossy, 3D, Arnold render, Vray
render, artstation, 4k, very
detailed, post processing”

12
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
„Well lit Nike pace shoes made of rubber and
composite fibres, black Nike logo in the shoe,
photography by Steven Meisel, beautiful
lighting, indirect ighting, unreal engine, octane
render, hd, iq 180 --ar 2:3 --q 2“

„A stunning interpretation of nike shoe


sneaker made of petals, advertsiement,
solarpunk, highly detailed and intricate,
golden ratio, very colorful, hypermaximalist,
ornate, luxury, elite, good vibes, oil painting,
cinematic, cgsociety, James jean, Brian
froud, ross tran --ar 9:21 –uplight“

“Nike shoes made out of branches,


leaves, flowers, moss, daisies, roses,
wood, in a forest, cinematic, photo
realistic, volumetric light. --ar 9:16”

13
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
1.4. Fantasy Illustrations

Now, let's get crazy. Let's create a few attempts for a trading card
fantasy game illustrations using DALLE - 2

“Fantasy ancient forest creature,


digital art.”

“Fantasy ancient forest creature. It is


misty weather and the creature reminds
deer, but its horns are made out of tree
branches, digital art.”

14
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“Fantasy ancient forest deer
and its horns are made out of
tree branches with leaves. The
deer is very detailed, he has
open mouth like it is howling.
There are colorful birds flying
around him. It is beautiful sunny
weather, digital art.”

“Fantasy ancient forest deer and its


horns are made out of tree branches
with leaves. The deer is very
detailed, he has open mouth like it is
howling. There are colorful birds
flying around him. It is beautiful
sunny weather, digital art.”

„Illustration of green dragon


made out of flora. The dragon is
in the jungle and it is sunny
weather. Little rabits are on the
ground a birds are flying around.
It is a design for trading card
game.”

15
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“Realistic fantasy illustration in the
style of "Magic the gathering" of green
dragon made out of trees, plants and
flowers. The dragon is in the jungle
and it is sunny weather. Little rabits
are on the ground a birds are flying
around. It is a design for trading card
game.”

“Fantasy dragon made out of


leaves, plants. flowers and
trees. Playing card illustration
style.”

“Phoenix in the night sky flying


through the mountains, fire is
spreading out of him, fantasy digital
art. The phoenix is very realistic and
detailed.”

16
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
2. RUN YOUR OWN AI

2.1. What is it about?


Now, when you have seen a few examples, it is time for you to give
it a try. In this chapter, we show you a simple way to run your own
AI software on your computer. All good AI tools offer you a short
trial, but sooner or later you will have to pay. From our experience,
those trials are really not enough to understand the way prompting
works, so it comes in handy to have one for free. We will focus on
a program called Stable Diffusion, which is an advanced tool for
generating images based on textual descriptions by using your
own graphical card (best suitable for NVidia cards) There are, of
course, many other AI- based image generators, such as the
earlier-mentioned DALL-E 2 and Midjourney, but Stable Diffusion
is the only free solution we know about that does not impose any
functional limitations. Thanks to it, even if you are a person
completely unfamiliar with programming or artificial intelligence,
you will be able to create surprisingly realistic images, all you need
is a simple text description.

17
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
2.2. Demo

An interactive demo is available on the official website of Stable Diffusion,


which allows you to try out the features of the program without having to install
it on your own computer. The demo is a great way to familiarize yourself with
the possibilities of the program and see how generating images based on text
descriptions works in practice.

To use the demo, simply visit the Stable Diffusion website and
find the "Demo" section. At the moment it is located between the
Model Access and License sections.

After clicking on the appropriate option, you will be redirected to


the page with the demo interface.

18
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Enter a description in the appropriate text field and click the
"Generate image" button. The program will process the text and
generate the corresponding image. You can experiment with
different descriptions and watch as the program translates the text
into visual representations. On this page you have some limitations
as to what you can do. There are some advanced options, but you
can only generate small images.

2.3. Download and installation

If you liked the demo version and want to get the most out of it, it's time to
download and install Stable Diffusion on your computer. The program is
compatible with Windows (10 and up), Linux, and within a few months with
MacOS. You will be able to use the full power of the program if you have an
Nvidia graphics card with at least 6 gigabytes of VRAM. However, if your
graphics card does not meet these requirements, the program will automatically
switch to CPU mode, although this may slow down the image generation
process.

Download the stable version of Stable Diffusion from the official


website. After downloading, unzip the file and move the folder to
the C drive of your computer. Note that this process may take some
time as the program folder weighs approximately 18 gigabytes.

19
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
After all is complete, open the Stable Diffusion program folder on
the C drive and find the file named "Stable Diffusion UI". Double-
click this file to run it. If you get a security prompt, feel free to click
"Run anyway" because Stable Diffusion is safe and trustworthy.
Don't be scared, a black window with white lettering, called CMD,
is a sign that everything is going according to plan. If something
interrupts the installation of the program at this point, please repeat
the double-click action on the file in the folder. The installation
should then continue from the stop point.

After launching the program, your web browser will automatically


open to the local Stable Diffusion server page, available at
“localhost:9000”. On this page you will be able to use the functions
of the program to generate images based on text descriptions.

20
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Remember, do not close the CMD window
while using the program!

2.4. Image generation

On the Stable Diffusion page, you'll find a text box where you can
enter image descriptions. Enter your description, for example “cat
drinking a beer in a bar”. Then below, you can adjust additional
settings such as graphic style or image size. After entering a
description and adjusting the settings, click the "Generate Image"
button. The program will start the image generation process based
on your description. You can watch the generation progress on the
program page.

Depending on the complexity of the image and the power of your


computer, the generation time may vary. After the generation is
complete, you will see the generated image on the screen. You can
save it on your computer and admire the result of your creativity.
In our case, we got such a result. Doesn't it look wonderful?

21
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
3.PageMaker
ANATOMY OF
including versions of Lorem Ipsu

A GOOD PROMPT

You've already discovered the power of AI image generation, and you probably
already have one on your own device! So now is the time to learn the anatomy
of a good prompt in detail. In this chapter you will learn what ingredients it has,
and in the next chapters you will find out why each of them is so important. After
all of that, you are going to become a prompt master!

Prompt anatomy ingredients with some instances:

1. Subject
2. Resolution – unreal engine, sharp focus, 8k, Vray
3. Lighting – studio light, dramatic light, open space, spotlight
4. Style – hyperrealistic, pop-art, modernist, art nouveau
5. Medium – digital art, acryl, drawing, photo, portrait
6. Artist – Alphonse Mucha, Frida Kahlo, John Collier
7. Website – pixiv (anime), pixabay (commercial stock)
8. Color – silver, vintage, pastel, color-splatter
9. Additional details – dramatic, low angle shot, godrays

22
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
4. SUBJECT
4.1. Subject definition

Now it's time to delve into the individual components of a good prompt. The first
thing we prompt is the subject itself. The most important thing is balancing the
amount of information we give to the AI generator, as being too vague gives us
too random results, but being too specific results into incomprehensible
hodgepodge, as the AI does not know which keywords are the important ones.

The correct amount of keywords and the optimal length of


sentences when providing a prompt for an image-generating AI
can vary depending on the specific AI model and its capabilities.
However, we can say that a good prompt should contain enough
information to convey the desired image but not be overly
complicated or confusing. Longer sentences can provide more
detail, but they can also be more complex and harder for the AI to
understand. In contrast, shorter sentences are often easier to
understand, but they may not provide enough detail for the AI to
generate an accurate image. Here are some examples of bad and
good prompts for generating an image of a warrior:

Bad Prompt: "Warrior, fight, weapon, enemy, castle, victory."

23
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
This prompt is too vague and does not provide enough specific
information for the AI to generate a meaningful image.

Better Prompt: "Generate an image of a female warrior holding a


sword and shield, standing in front of a medieval castle with a big
dragon in the background."

This prompt provides more specific information, including the


gender of the warrior and her gear. The AI also knows what should
be in the background. However, you can see that the generator
has no information about how should be anything positioned.

Good Prompt: "Create an image of a samurai warrior in traditional


armor holding a katana sword, with cherry blossom trees in the
background and a setting sun."

This prompt is concise and provides specific details about the


warrior's appearance, weapon, and surroundings. The inclusion of
the cherry blossom trees and setting sun uses the association
effect which will be discussed later.

24
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Bad prompt: “Create an image of a samurai warrior in traditional
armor holding a katana sword, who is looking prepared for a fight.
The katana sword is long and the warrior is swinging it. He looks
dangerous. There are cherry blossom trees in the background and
a setting sun. Birds are flying around the warrior.”

This prompt is overly complicated and the AI does not know what
to prioritize. We ask for a samurai who is looking like he is prepared
for a fight, but we also want him to be swinging his katana. Those
are 2 instructions about the change of posture and they may
contradict each other. The order of words is important and „Birds
are flying around the warrior.“ had been mentioned after the
instructions about the background and the generator did not know
where to put them, so the request was denied.

Better prompt: “Create an image of a samurai warrior in


traditional armor holding a long katana sword, who is prepared for
a fight. There are 4 small birds flying around the warrior. There are
cherry blossom trees in the background and a setting sun.”

25
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
For comparison, first series of images was generated by creative
fabrica, second with DALL-E. You can see that shortening the
prompt and positioning the request for birds close to the request of
samurai made them appear. You can also see that the
interpretation of the prompt is sometimes more random. That is
caused by the „randomness“ parameter of settings. Some of the
engines (like creative fabrica) do not let you change this, other
ones (like midjourney or stable diffusion) will offer you multiple
settings (stylize; chaos; seed; randomness; …). The randomness
for the second image generate by creative fabrica was high, so the
engine decided to also 4 samurai, not just 4 birds.

You can see that the generator decided to use a lot of style known
as ukiyo-e. That is due to the association effect. Most of „samurai
paintings“ that were in the dataset used to train this AI probably
were in this style, so it naturally becomes AI’s first choice. For this
reason, it is necessary to add styles and mediums to your prompt
and you are going to read about all of that in next chapters.

Simpler and free programs tend to generate output based on the


first keyword. Better ones usually give similar values to all of the
keywords and try to anticipate what would you find more important.
You can adjust the importance by commands. That is called text
weight or key strength. For explanation of text weight, we are going
to use Midjourney. Midjourney uses a lot of stylization
automatically, so even very simple prompts can give us amazing
results.

26
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
The following image was
generated only with:

“/imagine samurai and bird”

Now we can adjust the


command so both keywords,
samurai and bird have exactly
the same value. In Midjourney,
we write:

“/imagine samurai::1
and bird::1”

You can see that as both


keywords have the same
value, AI is going to mesh
these concepts together.

27
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Now we emphasize that we
want a samurai and the bird is
secondary, so we double the
text weight of keyword
samurai:
“/imagine samurai::2
and bird::1”

It makes sense that Midjourney is not sure about if we mean a


picture with a samurai somewhere and a bird somewhere else or
a picture where is a someone who is a samurai and a bird at the
same time as we did not give it enough information.

If we now emphasize the


samurai even more, the
keyword bird becomes almost
ignored:
“/imagine samurai::3
and bird::1”

28
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
We can easily reverse the
process and give more
strength to the keyword „bird“

“/imagine samurai::1 and


bird::2”

You can clearly see how much


more important it became.

Again, if we make keyword


„bird“ even stronger, the
keyword „samurai“ is going to
be ignored.

“/imagine samurai::1 and


bird::3”

29
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
We can change the value of any number of keywords in our prompt:

“/imagine Samurai::3 sitting


on a rock, bird::2 flying over
his head. Cherry blossom
trees::2 in the background
and a setting sun::3.
Japanese wooden
structures::1 are in the
distance”.

Any change of these values


seems quite impactful:

“/imagine Samurai::2 sitting on a


rock, bird::3 flying over his head.
Cherry blossom trees::3 in the
background and a setting sun::1.
Japanese wooden structures::2
are in the distance”.

Useful tools:
- Futurepedia
- Promptmania.com

30
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
4.2. Negative prompts

Negative prompts serve as a powerful tool in the Stable Diffusion


model, enabling users to specify undesired elements within
generated images without the need for additional input. By utilizing
negative prompts, users can guide the generation process and
exert greater control over the output image's content. Negative
prompts function as high-dimensional anchors within the
generation process, guiding it away from specific undesired
features. By introducing negative prompts, users can fine-tune the
output image with enhanced precision and detail. This approach
allows for the creation of unique images tailored to the user's
preferences.

To illustrate the application of negative prompts in Stable Diffusion,


consider the scenario where a user generates a portrait but
encounters duplicate images, despite specifying parameters such
as "Don't add duplicates." This occurrence arises from the model's
ability to comprehend negative prompts more effectively than
human language. Rather than repeatedly regenerating the same
prompt, users can introduce a negative prompt such as "duplicate."
By doing so, they effectively instruct the Stable Diffusion model to
exclude duplicate elements from the generated image. This
approach enables users to refine the output and obtain more
desirable results.

To further comprehend the utilization of negative prompts in Stable


Diffusion, let us explore a range of examples. These prompts can
be employed to remove specific objects or enhance the overall
quality of generated images. When incorporating negative prompts,

31
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
it is crucial to utilize words such as "exclude" or "there is no" to
indicate the undesired features clearly. Here are some examples:

- Exclude duplicates
- There is no overexposure
- Exclude bad art
- Without distorted face
- Exclude drafts

By incorporating these negative prompts into the Stable Diffusion


process, users can effectively manipulate the generation process,
resulting in images that align more closely with their preferences.

Here we have given above example. On the left, we have an image


generated from the "portrait of a woman with brown hair" prompt.
And the second time the image was made with the negative prompt
"There is no duplicates". You can see the result for yourself.

32
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
4.3. Keyword strength
Keywords play a pivotal role in picture generation techniques,
offering a powerful means to adjust the strength and bias of
specific elements within the generated images. This section
explores how keywords can be easily manipulated using
parentheses () and brackets [] to control their impact. By
understanding the concept of keyword strength and employing
blending techniques, users can create unique and visually
appealing images.

Parentheses (): Increases keyword strength by a factor of 1.1.


Brackets []: Decreases keyword strength by a factor of 0.9.

The weight assigned to a keyword determines its influence on the


generated image. The more weight applied, the greater the bias
towards that keyword. It is important to note that the effect of
multiple parentheses or brackets is multiplicative, with each
additional symbol indicating a 1.1 or 0.9 times increase or
decrease, respectively. For example:

(keyword): 1.1 [keyword]: 0.9

((keyword)): 1.21 [[keyword]]: 0.81

(((keyword))): 1.33 [[[keyword]]]: 0.73

33
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“Photo of eggs with bacon, on a frying pan”

(((bacon))) ((bacon)) (bacon)

Blending keywords is a valuable technique


for creating new and distinct looks within generated images.

This method is particularly useful when combining two


recognizable faces, as using the names of actors and actresses
can easily generate visually appealing results. However, one
drawback of this approach is that the resulting faces may be too
recognizable, lacking novelty.

To overcome this limitation, keyword swapping enables the


blending of two readily recognizable faces to form a new composite
face. The syntax for blending faces at different degrees is as
follows:

[keyword1 : keyword2: amount]

34
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
By specifying the names of the two individuals and adjusting the
"amount" parameter, users can blend the features of the two faces
to varying degrees. This technique allows for the creation of novel
and unique faces while retaining some familiarity.

The use of double colon „::“ is not only to add keyword strength, but it allows
you to separate different thoughts.

You will find out that language plays tricks on us and you will
sometimes need to explain your idea precisely to the AI.

35
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Here is an example:

“/imagine fire fighter”

“/imagine fire:: fighter::”

Now go on, try asking it for „hot dog“.

36
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
5. RESOLUTION

When you just experiment, you might want lower resolution as the render time
is shorter and it costs you less credits. But if you already found your perfect
prompt, you need to think about the exact resolution you want. Is the picture
going on a website? Is it going to be printed, imported to another app, or
adjusted in photoshop? Based on that, you might want to try different resolution
settings.

From now on we use just Midjourney as our superior.

Here is the basic prompt we are going to work with:

“/imagine An owl sitting


on a branch”

Now, Midjourney version 5 is out and all the images are stunning,
but it also takes a bit more of the processing power and if you are
experimenting, you might to want use older versions to save
yourself some credits. Let’s see what kind of results would we get
from other Midjourney versions. You can change the version by

37
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
adding command --v (and the number or name of version).
Currently available are all versions from 5 to 1 and also 2
experimental versions - a general purpose artistic mode you can
use by typing --test and photo-realism mode you can use by typing
–testp. There is also a „niji mode“ which is a collaboration between
Midjourney and Spell brush, it was trained on anime and manga
and is specifically designed to create this kind of Japanese art style.

An example of the same


prompt, but generated in
version 4.

“/imagine An owl sitting on


a branch, --v 4”

And version 3.
“/imagine An owl sitting on a
branch, --v 3”

38
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Version 2
“/imagine An owl sitting
on a branch, --v 2”

And finally, version 1


“/imagine An owl sitting on a
branch, --v 2”

You can clearly see why versions 1 and 2 are not really used anymore.

39
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
An artistic version.

“/imagine An owl sitting


on a branch, --test”

A photorealistic version.

“/imagine An owl sitting


on a branch, --testp “

And an anime/japanese art


version

“/imagine An owl sitting


on a branch, --niji”

40
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
People often try to use prompts as 4K or even 8K to achieve
extremely high resolution. Even when we write 4K, Midjourney is
not allowed to generate such a resolution, as it would take too
much processing power. Midjourney is currently able to generate
images up to 2048×2048 (square) and 2720×1530
(widescreen) using the beta upscale redo feature. The latest
version of Midjourney (V5) produces 1024×1024 resolution images
by default.

"/imagine An owl sitting


on a branch, 4K.”

Midjourney is not able to use 4K resolution,


it understands command 4K as „highly detailed“.

41
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
This is the upscaled version of variant 3 from the previous
generation.

An owl sitting on a
branch, --ar 4:3.
In Midjourney
version 5, aspect
ratio settings
completely replaced
–w (width) and –h
(height) settings that
are not possible to be
used anymore.

Here is an example of some of the resolutions you might want to


try. Now, when version 5 is out, you can experiment with any
resolution you want, but beware the fact that any aspect ratios
greater then 2:1 are experimental and may produce unpredictable
results. You can also see the available AR for different Midjourney
versions.

42
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Version Version
Version Version Test /
x 5 4a or niji
4c 3 Testp
(default) 4b

1:2 Only: 5:2 3:2 1:2


Ratios any* to 1:1, 2:3 to to to
2:1 or 3:2 2:5 2:3 2:1

Regarding the technical quality of an image, we might look at it


from 2 different perspectives: the resolution and the overall amount
of details. Some of the prompts you can use for higher amount of
details are „photorealistic“, „ultra photoreal“, „ultra detailed“,
„intricate details“, 4K/8K etc..

43
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine An owl sitting on a
branch, intricate details”

“/imagine An owl sitting on a


branch, photorealistic”

For lower details, you might have expected the opposite, as „low
detail“, „minimalistic“, „ultra minimalistic“, „one line“ and so on, but
as you can see, Midjourney is not really build for not being detailed.
But, there is another way.

44
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine An owl sitting
on a branch, low detail”

“/imagine An owl sitting


on a branch, minimalistic”

“/imagine An owl sitting on


a branch, one line”

45
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
If you want to play with the quality, you can work with the command
„--q“ or „--quality“. This command should affect the overall
quality of an image. The range is from 0,25 to 2, where 1 is the
standard quality. Lower quality images are faster done and cost
less tokens/GPU minutes.

Model
Quality .25 Quality .5 Quality 1 Quality 2
Version
Version 5 ✓ ✓ ✓ -

Version 4 ✓ ✓ ✓ -

Version 3 ✓ ✓ ✓ ✓

Version 2 ✓ ✓ ✓ ✓

Version 1 ✓ ✓ ✓ ✓

niji ✓ ✓ ✓ -

“/imagine An owl
sitting on a branch --q .25”

46
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine An owl
sitting on a branch -
-q 2”

Good comparison can be seen on following pictures generated


through version 4:

“/imagine woodcut birch forest --q _ --v 4”

--quality .25

47
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
--quality .5 --quality 1

48
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
6. LIGHTING
& RENDERING
Quite a big change can be achieved by adjusting rendering or
lighting.

Different kinds of rendering are easily


observed on closer objects:

“Nike Air Flight Huarache Shoe pride


month, bright, colorful, glossy, 3D,
Arnold render, Vray render, artstation,
4k, very detailed, post processing”

“Nike pace shoes made of rubber and


composite fibers, black Nike logo in the
shoe, photography by Steven Meisel,
beautiful lighting, unreal engine,
octane render, hd, iq 180 --ar 2:3 --
style 4a”

49
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“Realistic nike shoe,
cinema 4D,
ray tracing”

Arnold render is an advanced ray tracing renderer used for VFX


animation.

Vray/V-Ray is also a real time ray tracing system, but mostly used
for architecture or advertising.

Unreal engine is a videogame development tool, that has very


specific „almost realistic“ style.

Octane render is a GPU render engine that uses a way of


calculating final rendered images that aims to be photo-realistic.

Cinema 4D is a professional 3D modeling, animation, simulation


and rendering software. It is very popular for making elaborate 3D
models.

Ray tracing is a rendering technique that can realistically simulate


the lighting of a scene and its objects by rendering physically
accurate reflections, refractions, shadows, and indirect lighting.

50
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine Fantasy
medieval castle
on the mountain,
octane render”

“/imagine Fantasy
medieval castle
on the mountain,
unreal engine”

51
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine Fantasy
medieval castle on the
mountain, cinema 4D”

“/imagine Fantasy
medieval castle on the
mountain, ray tracing”

52
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine fantasy castle” - with different settings

Here you can easily observe all the differences.

If you want your picture for commercial purposes, try to play with
„studio light“ prompt:

“/imagine a can of soda” “/imagine a can of


soda, studio light”

53
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Different „light“ prompts change the overall atmosphere of the
image in a deliquiate manner:

“/imagine Fantasy
medieval castle
on the mountain,
volumetric lighting”

“/imagine Fantasy
medieval castle
on the mountain,
cinematic lighting”

54
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine Fantasy medieval
castle on the mountain,
softbot lighting”

“/imagine Fantasy medieval


castle on the mountain,

glowing lights”

“/imagine Fantasy medieval


castle on the mountain,
blue lighting”

55
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine Fantasy
medieval castle
on the mountain,
fairy lights”

“/imagine Fantasy
medieval castle
on the mountain,
long exposure”

56
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
7. STYLE

There are several commands you can use that will noticeably change the
stylization of the image. You might want to use them if you are looking for
inspiration or are not sure about the exact form of output you want. The
commands you want to play with are: „chaos“, „seeds“ and „stylize“.

7.1. Chaos
The parameter "--chaos" or "--c" determines the level of diversity
in the initial image grids. When set to a high value, it generates
unique and surprising outcomes with unconventional compositions.
On the other hand, lower
values of "--chaos" yield more
consistent and replicable
results.

“/imagine An octopus wearing


a hat”

57
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine An octopus
wearing a hat, --c 25”

“/imagine An octopus
wearing a hat, --c 50”

“/imagine An octopus
wearing a hat, --c 100”

58
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
7.2. Seeds
The Midjourney bot employs a seed number to generate a field of
visual noise resembling television static. This noise serves as the
initial foundation for generating image grids. Each image is
assigned a random seed number, but you can also specify a
particular seed using the "--seed" parameter. By utilizing the same
seed number and prompt, you can expect similar final images to
be produced. That comes in handy when you find the style you like
and just want little adjustments.

If you want to see the seed of your picture, just right click it and add
an „envelope“ reaction. The seed and „job name“ (unique code for
each image generated) will be shown in your chat. The seed
command accepts only while numbers in range 0-4294967295.
Yes, it is weird. The seed value only influences the image grid (the
blurry initial image you see during the generation). Without using a
seed, every generation has random one and the results will differ:

“/imagine An astronaut holding a banana”

Generation 1 Generation 2 (same prompt)

59
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine An astronaut holding a banana, seed 726”, And again

7.3. Stylize

This Midjourney Bot has been trained to produce images that favor
artistic color, composition, and forms. The “--stylize” or “--
s” parameter influences how strongly this training is applied. Low
stylization values produce images that closely match the prompt
but are less artistic. High stylization values create images that are
very artistic but less connected to the prompt.

Different Midjourney versions have different stylize ranges:

Version Version Version Test /


niji
5 4 3 Testp
Stylize
100 100 2500 2500 NA
default
0 0 625 1250
Stylize
– – – – NA
Range
1000 1000 60000 5000

60
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine A dancer in rain”

“/imagine A dancer
in rain --s 300”

“/imagine A dancer
in rain --s 600”

61
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine A dancer in rain --s
1000”

“/imagine An apple --s 8”

“/imagine An apple --s 800”

62
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine colorful risograph of a fig --s X”

“--s 0” “--s 50”

“--s 100” (default) “--s 250”

63
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“--s 750 “ “--s 1000”

64
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
8. MEDIUM
8.1. Specify the medium

The need for specificity doesn’t end when you fully and precisely describe the
subject of your desired image. You should also specify its medium.

Medium is the material used to make artwork. Some examples


include illustration, oil painting, 3D rendering or photography. In
this chapter, we will explore many more options including the visual
examples. You will understand how strong the effect of medium is,
since a single simple keyword alone can dramatically change the
visual.
Let's start with a few examples of art mediums with an elegant
maine coot cat.

„/imagine a Maine Coon Cat


sitting on a wooden table”

65
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Midjourney aims to provide the smoothest user experience-
therefore without specifying the medium, it uses realistic
photography as a default medium in most of the cases. We can
adjust a few qualities: the color palette, lightning and resolution.

„/imagine a Maine Coon Cat


sitting on a wooden table, soil
colors, studio lighting, high
detail”

It's time to include our preferred art medium:

„/imagine an oil painting of a


Maine Coon Cat sitting on a
wooden table, soil colors,
studio lighting, high detail”

66
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
„/imagine a lithography of a
Maine Coon Cat sitting on a
wooden table, earthy soil
colors, studio lighting, high
detail”

„/imagine a watercolor
painting of a Maine Coon Cat
sitting on a wooden table,
earthy soil colors, studio
lighting, high detail”

If you have enough earthy soil shades, let's add some color and
life!

67
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
„/imagine a crayon
illustration of a Maine Coon
Cat sitting on a wooden
table, vivid color palette,
studio lighting, high detail”

8.2. Sketch

As we already pointed out in the Subject chapter, the word choice matters.
For instance a word "sketch" can have multiple meanings depending on the
context.

A sketch can refer to a rough or unfinished drawing that captures


the basic outlines or features of a subject. Sketch can also denote
a brief or basic outline or plan, typically used as a starting point
for a project or design. For example, an architect might create a
sketch of a building to illustrate the general layout before
proceeding with detailed blueprints. However, as we can see in the
following examples, Midjourney will most likely associate the word
“sketch” with the first connotation- a simple drawing.

68
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine a sketch of a tomato”

As you can spot, Midjourney


offers you 4 more or less
various pencil sketches…

“/imagine black and


white sketch of tomato”

Bear in mind that the


importance (weight) you give to
each of the words also matters:

„/imagine tomato::1 sketch::"

69
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Unless you specify it, for instance with more details, like a

• Painting: Techniques for creating depth, texture, and color


harmony.

• Sculpture: Methods for working with various materials and


shaping three-dimensional forms.

• Printmaking: Introduction to techniques such as etching,


lithography, and screen-printing.

Watercolor sketch

“a boy and a girl, Dream


journey, dream world, enter a
fantasy world, colorful nature,
world of wonders, no letters,
watercolor sketch, v 5.1 --s
250 –“

Sketch

“/imagine black and white sketch


of food”

70
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
8.3. Combining mediums

There are no more limitations to our creativity. So what about


mixing traditional and digital techniques to create unique artworks?

“Vincent's Choice Digital Watercolor Paint of a Sunflower vector


art illustration 8k --ar 18:12 --v 5 --s 750 --q 2”

71
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
9. ARTIST
9.1. Artists style

This category involves referencing a specific artist or their work to evoke a


particular style or influence. Examples include mentioning artists like Alphonse
Mucha, Frida Kahlo, or John Collier.

Now, imagine you are a passionate writer investing your entire free
time into the creation of a breathtaking novel. You would like to
share it with a wider audience, yet you lack talent and money for
appropriate visual accompaniment inspired by your favorite artist.
Why not ask AI for help? You will be surprised by its “knowledge”
of the vast amount of artists, not only world-famous, but also those
known only in specific communities (e.g. anime or comic book
artists).

“/ imagine a cover for a sci-fi novel set


on a distant planet in the style of
retrofuturism, referencing the works of
Syd Mead. Use a digital art technique
and a 4:3 aspect ratio. The title is
'Aeronian Odyssey' and weight is 5.0”

72
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“/imagine a cover for a horror
novel set in a haunted mansion
in the style of Gothic art. Use a
pencil sketch technique and a
3:2 aspect ratio. The title is 'The
Darkened Halls' and the weight
is 5.0.”

“Innsbruck buildings urban


sketch by Albert Kiefer”

“:/imagine a cover for a mystery


novel set in Paris during the
1920s, using an impasto
technique in the style of Vincent
van Gogh. The title is 'Midnight
in Montmartre.' Use a 9:16
aspect ratio and a weight of 5.0.”

73
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
“urban sketch of the skyline of Rome, --ar 2:1 – “

9.2. Narration

Prompts can remain very simple or quite complex. Keep in mind that short
prompts rely heavily on Midjourney’s default style, while longer prompts allow
you more customization.

74
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
However, super-long prompts aren’t always better. Focus on the
main concepts you want to create.

“Create a digital painting of a


majestic mountain landscape in
the style of Impressionism,
inspired by the works of Claude
Monet. The artwork should
resemble the art found on
ArtStation, with a high resolution
of at least 4K. Use vibrant colors
and a serene atmosphere,
incorporating a pastel color
palette. Illuminate the scene with
soft, diffused lighting to create a
warm glow.”

Compare it to a result of a non-narrated, well structured prompt:

“:imagine a mountain
landscape, a digital painting,
impressionism, by Claude
Monet, ArtStation, 4K, vibrant
colors and a serene
atmosphere, pastel colors,
soft diffused lightning creating
a warm glow”

75
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
10. IMAGE PROMPTING

If you want to build upon the image you already have, or want Midjourney to
draw inspiration from specific image, you can use your own image or picture
from the web as a prompt.

To add an image from web, just paste the URL into the chatbox. If
you want to use your own, upload it via the plus symbol into the
chat and when you want to use it, drag it into the prompt. The
importance of the image compared to the text prompt can be
adjusted by command “—iw” (image weight) in range 0.5 to 2.
Lower number means lesser importance of the image and vice
versa.

Here we have our image:


https://i.sozcucdn.com/wp-content/uploads/2023/05/16/23-
erdogan.jpeg?w=1200&h=675&mode=crop

76
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
And our prompt:

“/imagine //LINK// as a sci-fi::3


alien::2. He is on a
spaceship::1 --s 350 –iw 1.25”

You can also easily mix images


together. Here we have 2
images:

Ship:
https://image.invaluable.com/housePhotos/weatherham/42/72304
2/H21808-L288079953.jpg

Sea:
https://media.istockphoto.com/id/626269490/photo/endless-sea-
and-sky.jpg?s=612x612&w=0&k=20&c=zJw-
UuIdRoS7jySKj_xw402eOYcMcMRh_OpQNLz-UXA=

77
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
And we merge them with an easy prompt:

/imagine A ship like this


//SHIP LINK//
on an open sea like this
//SEA LINK//

It is necessary to describe what should the AI do with those


images as just writing „+“ leads to beautiful, but very random
images:

78
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Bust:
https://cdn.shopify.com/s/files/1/0083/2471/0466/products/30-
5333_1_78ee2dd5-a97a-4490-a98f-
9541831e3bbe.png?v=1676450329

Fruits:
https://img.freepik.com/photos-premium/mangue-kiwi-figue-fraise-
raisin-poire-orange_2829-18589.jpg

79
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
11. EXTERNAL SOFTWARE
If you would like another app or AI to write the prompt for you, there
are some options.

When you try using ChatGPT, it is a long way to prepare it for this task. If you
just ask for the prompt directly, it gets creative, but the final result is not usable.

You can have a look:

If you really insist on using


ChatGPT, you need to feed it
specific information about
how to write prompts for the
AI you chose. Very good
explanation can be found on
website:

https://allthings.how/how-to-
use-chatgpt-as-a-prompt-
generator-for-midjourney/

Generally, you need to know


all the prompting rules for the
generator you are using and
teach them ChatGPT. You
can speed up the process by
showing it multiple good
prompts and let it ask what

80
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
different commands mean. It is not much efficient if you just want
to save time, but if you are planing to generate a huge amount of
pictures, ChatGPT could be handy. Luckily, someone already did
the job for you, so you can use chrome extension
„AIPRM“ (https://chrome.google.com/webstore/detail/aiprm-for-
chatgpt/ojnbohmppadfgpejeebfnmnknjdlckgj), which is ChatGPT,
but already fed with all the info about prompting.

81
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
If you want to get an
inspiration for
perfect prompts, you
can easily browse
this large „Prompt
gallery“

https://hero.page/samir/all-prompt-libraries-in-one-
page/midjourney-prompts-with-examples?ref=allthings.how

But if you really want to speed the process up and also learn
different commands, „promptmania“ is the way to go
(https://promptomania.com/stable-diffusion-prompt-builder/). It
allows you to choose the AI you use, then you write a basic short
prompt like „a Cottage in forest“ and you can add different
commands by simply clicking on offered settings.

82
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
It is able to use wide range of art mediums, color palletes, lighting,
post-processing and it has a huge database where you can choose
from dosens of styles of different artist.

It is a helpful tool for learning the basic prompt mechanics, but if


you have very specific image in your head, you will still need to
learn to do the prompting by yourself. At least for now … There is
also a lot of people on fiverr.com that are offering to write you a
good prompt for money, which seems completely ridiculous, as
there are tons of good free prompt generators already, but it is also
a possibility.

83
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
If this is not enough, you can get a lot of inspiration form the
„Huggingface“ (https://huggingface.co/spaces/doevent/prompt-
generator), which offers you tons of prompt variations.

Or you can try different prompt builders, as IMI PORT


(https://www.imiprompt.com/builder), which is very well executed;

Viorel Spînu, (https://www.viorelspinu.com/p/midjourney-prompt-


generator.html?ref=futuretools.io) which is simple, minimalistic and
easily comprehensible;

84
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
Or if you like
spreadsheets, you will
love Reddit projects
called „Prompter“,
which is a tool created
in Google Sheets. You
only need Google
account. It is not as
good as the others, but
is magnificent that the
community put so much
effort into this.

(https://preview.redd.it/twq07m5lnzb91.jpg?width=2048&format=p
jpg&auto=webp&s=ffa18bc96ef1ab9ee7b0c0a1cd2a75b803ef58d
e),

85
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
12. BONUS
“/Imagine Midjourney talking to ChatGPT”

86
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved
"When you close this book and the last page turns
over, remember that this is only the beginning.
Armed with the knowledge of using artificial
intelligence to generate perfect images, you become
the architect of your artistic world. Develop your
skills, experiment and discover boundaries that were
not there before earlier. Let your works become a
bridge between human creativity and the powerful
possibilities of technology. Now it's time for you to
take up the challenge and paint the future with the
help of artificial intelligence. Let your paintings
inspire the world and open new horizons of art."
~ ChatGPT

The Promptbook ®
Istanbul, 2023

87
USING CHATGPT AS A VIRTUAL COLLABOLATOR
Course final project, 2023 - All Rights Reserved

You might also like