Transcripts
1. Introduction: Dolly Three is one of the
most efficient platforms to bring your creative
visions to life. It generates professional
quality imagery based on the prompt that you give it in a matter of second. Hi, I'm Arnold. I'm
a creative director, educator, and a professional
in creating multimedia art. In the last year, I've been
researching and diving into AI applications for
professional creatives. And throughout the course
we will be learning how to craft efficient
prompts with dolly three. Understanding the
AI's capabilities to aligning with your
creative vision. Dissecting various styles and themes to make your
artwork stand out. Exploring the endless possibilities
of generative imaging and systemizing your
creative process using AI technology. This class is for
anyone interested in implementing AI into
their creative process. From artists who are just
creating for the sake of it, or even a freelancer whose content creation
needs some streamlining. For your final class project, you'll create your own
beautiful generation based on our workflow. And by the end of this class, you'll have a
professional approach to using dolly three for many
different applications. And a streamlined process
to master prompt writing. So if you're ready
to get started in the exciting world of dolly
three, let's get generating.
2. Project: Okay, welcome to the
project section. Now, the project for this class, we're going to help you
create something with the prop structure
that we are going to talk about in this class.
Now it's very simple. All you have to do
is follow along with that prop structure and create a work of art that is fitting
for what you want to make. Dolly is very powerful
and it's able to just focus on the specific
details that you give it. So really get creative, really think about all the different elements
that you can make. And then together
through this course, we're going to have
you come up with a beautiful submission
for your class project. Okay, let's go on to
the next section.
3. Accessing Dall-E 3: In this section we're
going to learn about how to access dolly three. Now there are two ways for you
to go about enduring that, and we're going to go
through it right now. So let's jump in. Okay, so
we're going to go in and talk about the different ways that you can access dolly three. Now, there are two
main ways to do it. I'm going to go through
the easy one first, which is using Microsoft Bing, which works with dolly three. So right here you see it's
powered by dolly three, and it's super easy to access. All you have to do is to go
on Bing.com slash Create. Or you can search Bing Image Creator on just about any platform
that you use to search. So with Microsoft,
Bing's image creator, you can go into the end phrase. And you see this here. This is just a
showcase of what they have going on and just some
frequently asked questions. The main thing that
concerns us is here. This is where we
enter our prompt and create what we
want to create. Now a couple other things
to note right here as well is this is a boost number. This lets you create faster. And when you run out of boost, then you have to wait longer for your
creations to be made. Right now, it takes
really quick, like 30 seconds to create
a series of images, but once you run out of boot,
you'll have to wait longer. And that depends on how
busy the server is and how prioritize your E or query is
going through in the queue. Now you could also
press Surprise Me here for a surprise generation. Or create if you want to
create what you have in here. So once I press Surprise, it gave me this little, an astronaut dance party on
the surface of the moon. Digital illustration. And then I can press Create. And then it's going to
render that out for me. So as you can see
as I press Create, it's going through and generating my image right
here on the bottom, you can see the progress bar on the right is where my
old generations are. So if you scroll down, you can see some of your
old generations. And you can click on it and then see your prompts as well. But right now our
focus is on this. These are the ones
that we've just created with the prompt
that we gave here. And if you click
on the As images, you can share it, save it, download, or customize. Now customize is cool, because it is kind
of like Canva, where you can go and customize a design or graphic
based on this image. Now let's go on and take a look. I'll show you what I mean.
So once you're in there, you'll see that this
is another box for you to write down
something so the AI can generate something that is a better fit for this image. Now you'll see on the right, there's some ideas on what they have for this
image that we made. And it's cool because you can maybe use this for a poster
or something, for example. We'll click on this and
then customized design. And then you can see this little animation bit coming to life. You can add your text here, maybe it's a space party and then fall 2023 or something,
something like that. But essentially it's cool because you could
move all of this to a design platform all
with a click of a button. So let's go back and we'll go onto our images and
look at these images. Now there's also another
one that I use called save. And I'd like to save
my different images to a different collection
because that way I can easily reference the
prompts that I had used. Okay, now let's go on to the other way to
use dolly three, and that is to go on to open
AI site which develop dolly. And you can click Try Chat GBT, dolly three, or you could
directly access chat GBT. And then if you're a plus user, you can go and press the
GBT four on the top, and then press on the
dolly button here. Now I just want to
note that if you're on the open AI site
and you click on the tri chat GBT dolly three, It's also going to lead you into this interface right here. So once you're here with
this chat generation thing, you are essentially open to generating any
image that you want. So you can have a comic strip
of a cat talking to a dog. So I'll run this just so you can get an idea of
what we're making. And after you run it, you'll see this little box pop up
saying, creating images. Now if somehow you want
to stop the generation, you could click Stop right here. But it's going to
take a minute or so. It's a bit slower than bings. But I think that's because we have way more generations that we can do on chat GBT versus
just doing it on Bing. And here we have
a comic strip and it's kind of funny even
though it's not perfect. You can see it's
not perfect since it's still developing
technology. But you can see that it does tie in some
concepts together, like the cast asking
the dog while you chase your tail and somehow the
dog has a laser point. So of course this
is not perfect, you have to run through it
and edit it and whatnot. But we have here
and it just shows the capabilities of what
this generator can do. So as you saw, there are
two ways to access Aly. One is through this and the
second one is through Ing. And the advantage is that
being one is faster but you're capped based on the number of boosts
that you have. Meanwhile, chagiBTs
is not capped, but the tradeoff is it's
going to be a bit slower. All right, so you should
have learned that there are two ways to get
to Ally three and we're going to now
go into how to prompt and the way
that you should think about prompting
when you're prompting and creating
your images on AI.
4. Best Practices & Tips: Okay, in this section
we're going to talk about the best practices and tips for you as you're
prompting on Al. Now there's a misconception
that if you go on a you can just prop the
best art works right away. And this is essentially
what sparked this huge debate on whether AI artists are
real artists at all. And the thing is,
there's actually a whole process behind
prompting so that you're going to be able to make
better prompts that better understand what you're
trying to create if you follow these
best practices. So let's begin. And the first
practice that you should have in all of your prompts
is to be clear in detail. So as you're writing
your prompts, the more clear and
detail you are, the more the language
learning model the AI is going to understand
what you want to make. And if it better
understands what you're going to make or
what you want to make, then you're going to make
better looking images. Because a lot of
times what we have in our head is not going to
be what the AI sees like. We're too general with what
we're trying to generate. So we need to be able to understand what
defines the scene. So for example, if
there's a beach scene, what is in this beach scene? Is there a car? Is there
sand on the beach? Is there rocks on the beach? Is there a mountain on one side? Is there a highway? All of these need to be taken
into consideration. Now there is a caveat, and that you should not overload the AI model with a lot of details because
that might confuse it. So make sure you have
enough detail and understand what elements
make for this image so that you only give it the
essential elements that make this image pop or make this image into what it
is that you want to make. So be clear and
precise, and detailed, but also understand what is most important so
that you can put it in there and include it and not confuse the
AA with too much. Now, the second thing
that we need to practice is to do
iterative prompting. And that means that you need to build on top of your prompts. So iterative means
having iterations. And iterations is having many different
versions of a prompt. So for example, if you prompt
something most likely the first time you've
prompt it is not going to be the best prompt. But if you keep building
on top of that, you're going to have a
prompt that's more closely resonating with the work of art that you have in your head that you want to bring to life. And after all, isn't most things in life built on top of another? Like if you think about this
house that I'm living in, there's bricks in the wall, and then there's
the mortar and then there's the layer that they put on top and then the paint. And then finally, they
finish this off by decorating the whole place
with interior design. And then if you want
to go even further, there's the branding element
to this whole place. So if you do
iterative prompting, you're going to be able
to better build on top of the initial prompt into getting what it
is that you want. Because a masterpiece actually
involves a lot of elements and you need to refine certain elements to
finally get to the end. Like the end product, the goal that you want to generate is going to
take a couple iteration. So it's good to do
iterative prompting. Now the next it is to experiment
with a prop structure. Now we do have a structure that we talked about
in this course, but sometimes if you flip
some of those things around, it might end up giving you a very different
creative output. And depending on what it
is that you want to make, that might really influence
how the product cars out. Say for example, I was
making comic strips. And if I were to put comic
strip and then three panel, it turned out a certain way. But then if I did three
panel comic strip, it turned out another way. So the more that you experiment, the more that you know
how your prompt is going to interact
with the AM model. And when you do that, it's going to give you the final product that
you want to make that's more closely resonating with what you have in your mind. And the fourth tip is to
simplify complex ideas. So say you have a really
complicated concept of an ancient town with
certain elements to it. And ultimately your
goal is to make this just a busy town
with a lot of life. You could just
simplify it by saying bustling ancient city somewhere. And that it's going to generate and that it's going to
give you something. And then you might like
it or you might see that, okay, let's make a new
iteration on this. And I'm going to add a
certain element that I want. So you're going to add
maybe a bustling city with a food market vendor. And then from there you add a food market vendor
to it. Because if you get too complicated with these prompts and make it
like this complex idea, many times the prop might not be able to communicate well with the AI learning model to create something that is a good fit for what you have in your mind. So as you're thinking
of your prompts, think about it in
a way that, hey, how can I make this more simple? And the more simple you're, the more clear and concise your, the more it's going
to be able to create what it is that
you want to create. And then finally, the fifth tip. The last tip is to practice. The fact that you're
practicing makes you more familiar
with the AI model. So as you practice and as you prompt more,
you're going to learn what certain prompts do, what certain prompts generate. And then you're going to
instinctively know like, okay, I want to
create something. These are the words
that I'm going to use. These are what I'm going to add. This is how I'm going to get rid of certain elements
that might pop up. And then eventually you're
going to know and be so fluent with this AI image
generation tool that you're going to have an idea in
your head and then fluently communicate with the AI model to generate what it
is that you want. The biggest struggle
with beginners is that they don't understand
what the AI model can do. So they just put out very
general generic type of prompts because they haven't had the practice to really think
about the capabilities and how to use the
tools and then how to appreciate all the little
intricacies of this tool. All right, that's the
end of this section. You should have learned the
five tips that you could implement and start practicing on your AI prompting today. Next up, we're going to go into the real time practice
of generating images, like the workflow
behind all of it.
5. Golden Rule of Prompting: Okay, so let's talk about the golden rule of prompting,
at least for Dolly three. Now, as we're creating
images with Dolly three, the most important thing
for you to keep in mind is to be as
specific as possible. Because Dolly three
interprets what you say quite literally and it generates something based
on what you give it. Now it also does have its
own creative control, but it's not like
the other platforms that just go very
wild, you know? Like if you want something
to be created and crafted in a way that
you have in your head, like the idea that you have. You need to be specific. Now that means that you have to think about many of
the little details. And this comes with experience. So it's okay if what
you do initially isn't as good as what you
see other people do online. And when I talk
about the details, I'm talking about
every single detail, the colors, angle, the way
that the character is framed. All of this comes together into a image that is cohesive with the prompt
that you give it. And so what this really
ends up looking like as you're working through
your process or workflow is that you might give
Ali a prompt and that it gives you a result back that you're not
very interested in. Like you're like, okay,
this is just alright. So from there you
have to go back in and be more specific
about what you want. So say for example, I'm creating a Japanese
anime style image. And I just say that and
add a character and then it creates something for me and then I see that I'm like, Al, right from here I think we can go in this
other direction. So then I add some
more elements. On top, I think about
maybe the lighting, the way that things are framed, the type of clothes
that they're wearing, what they're holding, what
elements are around them. All of these really go into
creating this entire scene, this images that you're making. So ultimately, it's up to you to create the amazing images
that you want to come out. Because you have to think about all these little
elements that make for an amazing piece of AI art. Now with that in mind, let's
move on to the next lecture, where we'll talk about
the prompt formula. As in a formula for you to think about when you are prompting
your art generations.
6. Prompt Formula: In this section,
we're going to learn about the prop formula. Basically, a formula
that you can use to think about as you're
creating your props. And with these props, it's a good basis
for you to start off of and think about how
to expand, add to it. Because the beauty about
AI is that it's not ad just limited into
this one prop formula. This prop formula is really
just a way for you to think about how to approach
generating images. Now it's pretty basic because I want to make it in a
way that everyone is able to understand
and get a grasp on how the prompting
works for dolly three. So once you do a few practices
using this prompt formula, think about how
you can expand on the different
elements that are in this prompt formula.
All right, let's begin. Okay, we're going to talk about the structure of
the formula now. So to break it down, the
formula really is core subject than descriptors,
than environment, or the settings, than the style or artistic
influence you have, and then any additional
details at the end. So let's go and start by
talking about the core subject. Now the core subject
should be first, because this is the highlight of your image that
you're generating. Unless of course,
the first thing that you list is something
very light, not something that
overtakes the whole image. Maybe it's just something like
daytime, then that's fine. But if you want it to be
a particular subject, for example, that should
be the first thing. So that when the
language learning model looks at what you wrote, then it takes that first word and it builds on top of that. So the core subject
is the main focus or the object of your piece of work and tips
on that would be, you should be clear about
the primary subject so that the generation focuses
on what you're making. So an example of this would
be like, oh, this is a car, this is a spaceship, this is a surfer at the beach. Just core subjects that will be the key highlight
of this image. Now moving on, we
have descriptors, which is something
that describes the qualities and
characteristics of the subject. So for example, this could
be futuristic or this could be a Neo Noir
or French inspired. Essentially things that would go along with the core subject that will be more specific about what
you're trying to make. And it's actually very
important because it influences the visual aspect of your
subject very heavily. And this is kind of a way
for how you can showcase your creativity because
the language model doesn't have the
creativity that you have. So if you have any
interesting ideas that are not something
that's typically found, then this is where
you may put it in. Like maybe you want to do a cat in an Eskimo suit
or an astronaut suit. Now of course, that
might be more common, But just for example, think of something that
you might not see in everyday life or
everyday artworks, and this is where you put your
own creative touch to it. Now the next part is
the environment or the setting that
this artwork is in. So this could be where you describe the scene
that's happening. So your character is maybe on a moonlit street or a urban city center,
or there at a beach. Now this is really just to tie everything
together because sometimes your image might come out with the generation
in a blank background. Now of course, if you
want a blank background, you can also add in that this is going to be on
a blank background, so maybe a blank
white background, say you're trying
to make a logo, or if you're trying
to make a character a sprite sheet for example. Then having a blank
background may be beneficial. So keep in mind that
having an environment, or a setting, or
a background for it is important for
you to consider. Now moving on, there
is the style that you want or maybe an artistic
influence to your image, because everything
that's generated is going to be super different. Imagine if you were trying to do something that is
more Western catered, You might do something
that's a comic book inspired by Marvel. Whereas if you're in
the Eastern world, if you're in Asia, you might prefer to do
something that's more of an animate type artwork because that resonates more
with the people there. Now of course, there
are definitely going to be interactions
between both sides. Like there are people in
the West that like anime, there are people in
the East that like comic books from Marvel, these are just examples. But essentially the
takeaway is that, think about the type
of art that you're making and this is where
you would insert it, so that it draws inspiration from these
types of art styles. And then finally, the section
for additional details, like maybe there's
something else that you want to add onto your image, but it's not like the core
subject of the image. So for example, you
could say like maybe there's rain falling
down on the scene, or there's fireflies surrounding
the whole background, or there's like a cherry
blossom tree or something. Basically just extra things
that you want to add onto it. A key thing to remember is
this is where you might add some flare or some depth to your piece just to make
it pop out and stand out. And a little tip is make sure
you don't go crazy with it. Maybe if you're doing it,
do it slowly step by step, like have one
generation see how it affects the image and
then add more onto it. Because if you do too many
additional details at once, it might confuse dolly and that might give you something
that's not what you expected. So essentially these are
the elements of a prompt. Now up next we're
going to go and talk more about how
to generate on dolly.
7. Generating Images: Okay, so this is
my favorite part, which is generating
the images, the art. So let's start by going
onto the new chat section. And if you clicked on from
the dolly three page, you would have landed here. But if you are already on
Chat GBT's user interface, then you can click New Chat. And then this is
what you'll see. Now if you go and
hover over GPT four, you'll see there's a couple of different options and on the
bottom there's dolly three. So with this, I'm going
to start by referring to the list that we made earlier of the formula for your prop. Now the first thing
is the subject. So I'm going to
have a pre female. And you could add a
little more detail in that by saying, pre
Vietnamese female. I am Vietnamese. So let's throw that in there
and then we'll add some descriptors to
this Vietnamese female, which is our second part. So I'll add descriptors such
as riding a Honda Dream, which is one of the most
popular bikes here in Vietnam, I'm currently in Vietnam. And then the next
part would be through the landscape or putting in the environment or
setting that they're in. So I would say
riding a Honda dream through the country
mountain side. So I added riding a
Honda dream through the country mountain sides and then I add it
in manga style, but I do like more
of a vintage feel, so I'll add in '90s manga style. Now this is where the
fun part comes in. Say I want to add a creative
touch to this manga style. And I know from my practice
that I've developed, or rather I've put together a sequence of words that gives
it a effect that I like. So I'll paste this in here. Now I put it aside on
the bottom so that it separates so we can read it and understand
what's going on. Now this is where the
extra details go. And from me iterating a lot of different
prompts and whatnot, I found this to be
something I really liked and this is what
it includes now. It includes using muted earth
tones and incorporating a grainy texture reminiscent of traditional celluloid film, which is some of the
techniques in old anime style. So this is going to
give it that look especially with the grain
and the celluloid film part. And then I also added it
should feature soft shadows in bold tory lines and
minimal color gradients. Now this is some of the
visual elements that will eventually give it the ultimate look that
we're trying to get. And then it finishes
off with saying, evoking a sense of nostalgia and warmth akin to
vintage anime series. Now this just locks it up and ties it in that this is going to be something in the style
of a vintage anime series. So finally, our prop is a pretty Vietnamese female riding
a Honda dream through the country mountain sides
in '90s manga style using muted earth tones
and incorporate a grainy texture reminiscent of traditional celluloid film. And the artwork should feature soft shadows, bold contra lines, and minimal color radiance, eventually evoking a sense of nostalgia and warmth akin
to vintage anime series. So let's run this, and
here is our render. So you can see there are many
different styles in here, but for the most
part there are a few that really hit what I want. Like this one, for example, has that old anime
type of look perfect. This one too was really cool, that old type of look, '90s kind of manga looking. But then this one is also a kind of a mix
between the two. It also has some
Vietnamese elements to it because the way that
the face is drawn is actually pretty similar
or reminiscent of a lot of the older
Vietnamese art that I see. And then finally, this one, which is very different,
it's super three D, but of course not all of them is going to turn out
the exact same. But this is really
cool. She's wearing a traditional dress and she's riding a motorcycle
through the mountain sides. So that's it for
my work process. And my favorite part is
really checking out what you make just to see how beautiful you can
create these images. And when I experiment, I like to change up
a lot of this part, the additional details part, because you could really create an entirely different scene just by changing this
part right here. You can make it into a
cinematic movie type, you can make it a poster, or you can make it
something really niche, like we just did with a vintage Anime feel that has a very nostalgic
feeling to it, especially for those who grew up in the '90s
that watched anime. If you want to make
modifications to this, for example here, I want
to do a different bike, so I'm going to make
it attempt that. And I typed here, make it a Honda Cub instead. And the beautiful thing
about this is because it's a language learning model and Chat GBT has one of the most advanced
language learning models. It's able to understand
what you're trying to say. And in this reference,
you're referring back to the image
you just created. So it's going to understand that this is referencing this. And let's see what
it comes up with. Now this is a good way to update or make revisions
onto your image. Because like we said earlier, iterations are
really important to how you can create the
masterpiece you want to create. Now, as somebody that knows
what bikes look like, the first generations did not come out with a Honda
dream that I wanted, so I instead asked for it to make another bike
that had a similar look, which is also another timeless classic bike in the country. And this example is more close to what I had
envisioned, This one as well, it's almost spot on, whereas the first one we had was actually
like a motorcycle, which not too many
people drove here. And then this one as well is very close to exactly
what I wanted. So as you can see, it's one very good at understanding
what you're trying to make because here
it's able to reference these motorbikes and create an image that is of
that exact motor bike. And so if you want to continue editing onto this image and
creating more life to it, we could instead add
a phrase like this. Add rice paddies to the background and
as we see here now, there are brice patties
in the background and the revision that we made
earlier is still present. So we can see the Honda Cub, which looks like the
Honda Dream Bike. This is also a very
similar Bike 12 and it has all the
elements that we wanted. Now there are some
suggestions here like maybe add a sunset backdrop
or a rainy weather setting. I'll go with the sunset just
to see what we can get. But essentially the takeaway
here is that once you build your image up by having
that first prop generate, if you build more revisions and iterations on top of that, you can get closer to this masterpiece that
you're trying to create. Because if I were to just use
a prop in the initial part, it might get confused and not include all of these
different elements on there. But as I build up on
top of it slowly, we're able to see how these
things start coming to life. So this is what the latest
iteration came up with. It's having a sunset backdrop, and this is what
we have somebody on the bike that we wanted, the rice patties, and
also the countryside. So as you can see,
super beautiful art. And the especially
cool part is that it's exactly the type
of style that we want. Okay, that's it
for this section. Up next, we're going to conclude the class and I'm excited
to see what you make.
8. Conclusion: Congratulations on
finishing the course. Now in this course, you should have learned
the many different things that go into prompting
on dolly three. The first is the thought
process on how to prompt and how to be specific in what
you're trying to make. The next is the parts of the prop formula or the structure of how
your prompt should be. Now remember, this
is just a guideline for you to start off with. Once you get more advanced, you can expand onto it. And then finally, we had some examples and a practice
for you to go through. Once again, congratulations
on finishing this course. I'm excited to see what
you make and make sure you submit it in the project section below so I can
take a look at it.
Arnold Trinh, Mixed Media Creator
