Transcripts
1. Introduction: I have made art in
the past as play. If I needed a break, if I needed to recenter
myself, I made art. Bringing AI into it is about using it as another
tool for tinkering. Hi, I'm Linda Dounia Rebeiz. I'm an artist and designer who investigates AI's
biases and blind spots. Not all AI tools are
created equally. It's important to be discerning about the models
you're going to be using and how they're trained in order to make
the ethical choice. This class is about the
big hairy questions that AI poses, and that might make us
very afraid as artists as it increasingly takes
over our industry. It is about engaging
with those questions critically and putting them at the center of what we make. We're going to start this class
by looking at how we make images and how that's different from how AI produces images. We're going to make portraits, environments, and ultimately, everything we make in this class is going to come together into a zine that will spell out
what we think about AI, and how we plan to use it. For me, this class is about reclaiming AI as a
method of production in our practice that can be used in a very
ethical and fair way. My hope is that by the
end of this class, you'll be able to
feel confident about your perspective on what AI is, how it produces images, and whether or not it
fits into your practice. I'm excited to explore some big hairy questions around AI together. Let's get started.
2. Making Images: In this first lesson,
we're going to be looking at how we humans create images. The reason we want to do
that is because we want to contrast that with
how AI makes images, which is a very
different process. To illustrate this, I'm going
to make an image today. As I'm doing this, I'm
going to talk about what comes into play
when I create something. I'm going to be using some materials that
I've made before, but also some materials that I'm going to be making in
this lesson with you. When I start to make an image, I always start with words because I like to
write my thoughts, how I'm feeling as
a starting point. I'm going to take
a piece of paper and then I'm going to start a mind map of the word home, which has been on my
mind a lot lately. Then I'm going to write down every thing that comes to my
mind as I think about home. Could be the present,
could be the past. There's my dad,
there's my daughter. It could be a place. This is where I used to spend Sunday during the
summers with my granddad, who then comes to my mind. We used to have a famous
rice and fish from Senegal. Are there images, because right now I have
a lot of people? I have some moments. Let's see. Can I bring up some places? There was a staircase
going up to my house. There is a snack
that I used to love. It's called Pido and it's
basically a sandwich with tuna, but it's very spicy tuna. I used to love that as a kid. Now I'm going to use a different
color to circle some of the words that are
bringing something into my mind that I want
to explore further. This staircase. Coco Beach was a hotel that we used to go to
during the summers. Actually, those
two are the ones. What I like to do
is use these as north stars and just play
around and see where I get. Let's just start with
the word Coco Beach, which is a place in my
hometown [inaudible]. I'm going to open maps, and I'm going to write Coco Beach Hotel to see if I can remember based on this map. This map shows all
these little squares and boxes that are the rooms. I think this would be reception. Now what I'm going to do is take a screenshot of the layout of the hotel and I'm then
going to use that as a starting point
to make an image. I'm going to take that and
send it over to Procreate and start making annotations on the actual photo of
the map of Coco Beach. What drew me to this image
were these little boxes. I'm going to redraw
them very loosely. They don't have to be
perfect, just like that. Then I'm going to try to
add some memories on top. I remember that we
would come through here, straight to reception. There were some trees
here all along the path. What else do I remember? The way to the
beach was this way, and right here, there was a little dance
floor, very random. Then I'm going to
keep doing this and just keep remembering stuff. Go back into that
childlike exploration mind from when I used to go there. Then what I'm going to
do is going to use this as a blueprint to actually
make the artwork. But the entire exercise so
far was about remembering, from the mind map to
making annotation on the actual map of the place
that I'm trying to remember, it's about conjuring up these
memories and emotions and then using that to
make the final piece , which I've already made. What I've done is that I've
taken different symbols from Coco Beach and I've
turned them into an artwork. There were these telephone wires when you walked in to the hotel just in front because there's
a big antenna nearby. I've attempted to recreate
them in Illustrator here. You'll see them in the
different compositions because of how striking the patterns of the interconnecting wires was, and it's something that
comes up to my memory. I try not to question it,
so I've just added it. The other thing was the plants I was referring to earlier. The reason why I use
Illustrator for this particularly is because I can
play with geometry a bit. That's vector based. I can add different
elements, different objects. I can use different techniques
like image tracing. For example, these
flowers are image traced with just keeping the outline and then coloring the inside. I can also play with
text overlaid onto illustrations which I've done in the second set of
images, for example. I also have some of
the words that I written in the initial mind
map for this exploration. I wrote in my notebook, I tried to hang on,
but home was gone. Walls are fading to
dust in our memories. I've integrated that into some of the sketches
I have as well. While I never really know where I'm going when
I start making something, when I started composing
the images and putting the different
elements together, I decided that these
would be posters. Some of these have
actually been printed and put on display
in my hometown. Here's one very near the
house where I grew up in. Here's one on the wall of my grandmother's
house, actually, and then here's one on a
construction site for a stadium. Now that you've seen
how I create an image, we're going to ask
ourselves questions about three specific parts
of the creative process. Let's run out these
questions as we go. We're going to start
with influences. What themes are
present in your work? For me, from what you've seen, it would be home and the
longing for my childhood. What inspires you to
start making an image? This is really getting at your emotional state
or your motivation. For example, for me, it's about nostalgia and feeling a sense of loss and trying to
find a center that really is one of my
biggest motivations for actually making images. How do personal experiences, identity, and emotions
shape your work. Shape is a big term. It could be loosely
inspiring your work. But personal experiences
for me play a huge role. Identity, knowing
who I was, who I am, knowing or being aware of the emotions that
come into my work, or sometimes creating to
process those emotions. But these are three
questions about what influences you to make art? As you'll see in
the next lesson, is going to be very hard
to answer for a machine. The second set of questions is about techniques and
the mediums you use. First question, how do
you decide what to use? In the little example
that I demonstrated, it really started
with looking at the actual map and wanting to do something that looked
a little bit like that or something that was
typographic, for example. Sometimes the work guides what you end up using or sometimes you use the same
medium all the time. It really depends on
how many disciplines or how many mediums you're
combining in your work. But asking yourself
this question is always very important for you to know where
your process does start. Second question, what are the first steps you
take when you start? Art doesn't always
start when we make it. It can start before. It can start maybe
the night before or the week before when you're imagining or dreaming up
what you're about to make. These are all things that inform you more on who you
are creatively. Third question, do you
work quickly or slowly? Do you plan or improvise? Obviously, these are not
straightforward questions. There's no right or
wrong answer here. You can be someone who
works very quickly and you know exactly
where you're going and you get
there really fast, or you could be someone
who takes their time. Depending on the medium I use, I do like to do a
bit of planning, which is why I favor using my journals to record my
process, for instance. The last one has to do with your process
and your workflow, so how you actually
do your work. First, what does
your workflow look like from idea to
finished product? For me, it starts on paper, it starts with the reflections, then maybe there's
some sketching involved, and then chaos. I open a software and it's
chaotic from that point, or maybe if I'm painting, again, the studio
turns into chaos. It really depends on you. But being aware of the entire workflow
is going to be interesting as we
look into Lesson 2. The second question
is, what role does intuition or emotion
play into your process? For some things, I can get
very process oriented. For some things, I will rely on my intuition, for example, knowing when to stop or
placing something on a page. It could be from something I've learned like color theory, knowing what color to use, but it could also be
just feeling it out and trying different things
and seeing where it takes me. How do you know when
a work is complete? For me, for instance,
it's never really about having an idea in mind and getting to the
finished product, but it's about exploring and feeling some end as I go along. These are nine questions
about technique, process, and influences that are really helpful for you
to understand what kind of creative you are
and to differentiate yourself from the kind
of creative AI might be. When we go into Lesson 2, we're going to be looking
at how AI makes images. It won't have answer to
these questions because it doesn't have a creative
identity in the same way we do.
3. Making AI Images: In this lesson, we're going to be looking at how AI produces images and why that's very different in terms of mechanics for how
we produce images. While we're going to be
focusing on diffusion models, I'm going to start with
an introduction of different models and how
they generally work. We'll also get to talk
about datasets and why they are important for
image production in AI. We're going to be using Midjourney to create
images with AI. There are other
equivalents out there. You could use Gemini,
you can use ChatGPT, you could use Stable Diffusion. They usually all
do the same thing. One important aspect of how diffusion models create images is something called denoising. To show you how that works, I'm going to open
Midjourney and I'm going to prompt for a cat; very simple. I'm going to prompt
for a cat in a tree. Then you'll see that the models start to do
something that is denoising. As you'll see here, the model
starts from an image that's completely random pixels
and then gradually as it goes from 80% to 90% to
100%, it becomes a cat. This is really important
to know about how AI makes images
because as we'll see, this is a result
of how AI learns. The AI started with
a random noise and gradually turned that into a real image and that's
not by accident. The way diffusion models
actually learn is by being given images and learning how to add noise into the image until the image is
invisible anymore. That's called forward process. They do that over and over
again and once they've mastered the ability to add noise and increase
the noise of an image, they then learn the reverse
process which is to decrease the noise of an
image in order to create the image in
the first place. Not all AI works the same. In the example we've
just been through, we looked at a diffusion model and we looked at the
denoising process. But you've probably used
ChatGPT before which is a large language model
or LLM for short. Or you've used a
model that helped you train Aurora before or
maybe you've trained a GAN. These are all different kinds of AI models that work differently. For instance, a GAN is
very different than a diffusion model
in that you have to provide the dataset
to train the GAN. But when you prompt
a diffusion model using Midjourney like
what we did today, you don't have to necessarily
bring in your own datasets. Because they work differently and they work
differently than humans, AI cannot be a substitute
for creativity necessarily. But if you bring it
into your process and you're aware
of the different strengths that
different models have, then it can inform a pretty
interesting practice. All these models
work differently. They are very different
from how we create images. AI starts with noise. It starts with a random
image and then it gradually refines that image to create what we've asked for. That's very different from how a human makes an image
because we start with intentionality and that is a fundamental difference between human versus AI-made images. The way we learn
is very different. AI learns in a very
procedural way. As humans, we learn
from experience, we learn from examples, sometimes we even
learn from osmosis; like just being around
something helps you learn it. For example, can you recall
how you learned how to speak? With an LLM like ChatGPT, we could probably write down exactly the steps that it used
to learn how to say stuff. Another fundamental
difference between how AI makes and how we make is that we usually
come with context. We come with a culture and
certain practices that inform the way we work
or the way we create. But AI usually only has a
training data to rely on. It will make predictions
and find patterns within that training data in order to create
something new. It doesn't necessarily
come with a context or a particular cultural lens that isn't present
within the dataset. Despite the fact that
there are different kinds of AI models that all
work differently, one thing that's
important to all of them is the data
they are trained on. This data or these datasets
are millions and millions of images that are
collected and gathered and labeled in order
to train the AI model. It's very important to
interrogate the datasets that are used to train AI because there's no
dataset that's neutral. A lot of the datasets that
we have right now are directly taken from what we've
put out on the Internet. As we know, the
Internet penetration is not the same everywhere in the world so some cultures might be more represented
than other cultures. This is evident when you
start to interrogate an AI. For example, biases in a dataset can be very obvious like when
you're generating a face. If you just ask for a face and don't necessarily
give the context, the AI might give
you more likely than not a Caucasian face versus a black and brown
face or it can be a stylistic and
very subtle bias. For instance, always
having words in AI images written in Roman characters as opposed
to other languages. In the previous lesson,
I had made a few images reflecting on home and trying to hang on to it.
Let's look at them here. I'm going to use one
of these images as a reference in order to
illustrate this next step. Let's say this one that says, I tried to hang on
but home was gone. I'm going to take that,
which is channeling this idea of creating based on nostalgia or using
nostalgia as a fuel and then I'm going to pop
it over into Midjourney. It would be interesting to
find out how differently AI would conceive of the
sentence that's on this artwork. I tried to hang on to home, but home was lost. Now that we've seen what
the AI has created, there are four
images that I don't particularly think
respond to the prompt. It's probably looked at
or thought about home as a physical object like
in this image where the house is flying away or; I'm not really sure
what this image does, but it's someone
climbing a tree. Then there's an old house that's a decaying house which is a little bit more poetic
than the other ones. When I was thinking
about this prompt and this we've seen in
the first lesson, I made something that looked
like this very abstract, but to me, feels very connected to the idea
I was channeling. What we see here is that
AI doesn't respond in the same way to something as abstract as this sentiment or
the sentence or this goal. It needs instruction. It needs to actually know
what you wanted to create. If I wanted to create with AI something that looked like
the images that I made, I would have had a
very different prompt. It's not able to process
things the same way we are. This is why in this exercise you're going to get
to do this again at home and compare the two images and feel free to play around
with different prompts. They don't have to be perfect
prompts. That's the point. They just have to
be things that are usually enough for you
to start creating. As you compare to the images, think about these two questions. How does the process of
creating an image with AI using any of the
prompts that you use differ from the
way you create? Did you find
anything surprising? For example, when I tried this, one of the images
included a cat probably because the previous image
I prompted was of a cat. Meet me in the next lesson where we're going to
be asking ourselves the question of why we want
to use AI in the first place.
4. Finding Your Why: [MUSIC] In this lesson, we're going
to be asking ourselves the question of why we want
to use AI in the first place. We're going to work
on a little diagram together that I hope to be a living document that we can revisit
after every lesson, just to make sure we are
keeping in touch with our feelings about AI as we
explore different tools. What you're going to
need for this lesson is very simple, just
a pen and paper. It can be as big as you want. If you want to use
Post-its to write your notes or to change
them, you can do that. For the purposes of this
lesson I'm going to use an A4 sheet of paper
and a Sharpie, but you can use a whiteboard, you can use a pin board, whatever makes
more sense to you. You're basically going to divide the page in five
different sections. This would be my bottom section, which I'm going to divide by two down the middle of the page. Then we're going to divide the top part into three
different sections. Now that we've divided our page into five different sections, we're going to go
ahead and write in the middle section
the word values. This is really about what values you bring into your practice. What are the things
you advocate for? What are the things that make you comfortable or uncomfortable about the world that you try to bring up
in your practice? Which values you basically
uphold as an artist? Now, to the left of your values, you're going to write
down motivations. This is going to be
specifically about AI. Why do you want to bring
AI into your practice? What is motivating
you to do this? Whether that's a
fear, a concern, or something positive, write them all down
in this section. Then to the right of the values, you're going to have hopes. What do you hope to achieve by exploring AI
into your practice? Is it about refreshing
your ideas? Is it about helping you
generate newer ideas? Now in this bottom quadrant, you're going to write concerns. These might be concerns
that you gather along the way as you
explore different AI tools. Then the final one is questions. You may not have all the
information you need in order to make a decision
about how to bring AI into your practice,
and that's okay. You have unresolved questions. You have curiosities that
are coming up along the way. Please record them here. Along with values,
this is probably the most important quadrant as it helps you keep in touch with what you can discover next, what you need to
read more about. Make sure to keep
this section alive. Now that you have
your five quadrants, you can fill them out now. It doesn't need
to be exhaustive. Just write as much as you can and then put it
somewhere that's visible, so that every time
you're taking the class, you have it as a reference point and you can add to
it along the way. I'm going to show you what
this looks like for me. Some of the values that
I try to bring into my work has to do
with visibility. I always wonder, what does it mean to be
visible in the world, what does it mean to be
invisible in the world, and what are the power
dynamics involved in who gets to be
seen and considered? That actually gets me
into these other values, which is about exposing power dynamics or showing
them in some way. Another value is
probably family. My family comes up
in my work a lot. I love them. I'm going
to put them there. These are examples of some of my values that I
bring into my work. An example of a motivation
for bringing AI into my work would probably be
related to one of my values, which is invisibility
and visibility, which is about making sure I feel represented
by this tech. That is something that's
very important to me, as I don't want to feel on the periphery of how AI evolves. I want to be in
the center of it, and I want to get the chance to determine it has a future
for people like me. An example of a hope
that I have with AI is that when used right, it actually has some pretty
amazing capabilities. For example, one of the hopes
that I have is that it's used more for
recreating the past, especially with a fairer lens. We know, for example, that
the past wasn't always recorded in the fairest
and most accurate way. I would love to see more
efforts in making sure people are telling their own
stories in the way that feels right to them instead of someone else
telling their story. Lots of concerns. I always
have concerns with AI because it's just capable of so much and it's being used for
a very specific, profit-oriented motive that
I'm not super aligned with. I'm going to write here that it's profit-motivated
at the moment. The biases are very
concerning to me. Lots of issues with privacy and how our data is
collected from us. We don't always have a say, or we're not always informed
when it's happening. Some questions that I
still have about AI is, is there anything beyond the current models
that we're presented? Are there experimental models
that maybe haven't garnered as much financial
support but maybe are ecologically more conscious? I'm going to write that as well. Is there a future
for ecology with AI? Is there a world in
which they coexist and one doesn't have to take
so much away from the other? This is an example of
me filling this out. Obviously, it's a very
shortened version, and it might take you a
little longer to do yours. But again, it doesn't have to be exhaustive,
just get started, write things down,
and then put it somewhere visible that you
can continue to refer to. This document is going to help me put together the final zine, which is going to be
my manifesto on AI. It's really going to
be your North Star, and if you keep updating
it throughout the lessons, it's going to be very
helpful for you as you spell out exactly how you think of AI, how you want to integrate
it in your practice, and which areas of
it still make you uncomfortable and that you
want to continue exploring. Go ahead, make your
diagram, fill it out, put it somewhere visible using whatever format makes
more sense to you. Meet me in the next lesson, where I'm going to
show you how to generate faces using AI. [MUSIC]
5. Generating Faces: [MUSIC] In this lesson,
we're going to be generating faces using AI. But really, we want to ask some pretty gnarly
questions like what does it mean to
generate a face with AI? How does it do that? Where
does the face come from? You've probably seen
horror stories of AI generating faces or deepfakes that are used for
nefarious purposes. In this class, we're
going to reclaim that and figure out whether there's a way to work with
faces and portraiture, in general, using AI in a way that feels a
bit more ethical. For this lesson,
you're going to need basic collage materials like a cutting knife, scissors, glue, and we're also going to need access to a printer
because we're going to be printing images that we're going to be
creating with AI, but you're also going
to be printing images of people that you'd like
to bring into this collage. Please do get their consent. When I first came across AI, it was through GANs, generative
adversarial networks. It was specifically StyleGAN2. StyleGAN2 was trained
on millions of images of faces that
were found on Flickr. One of the things about StyleGAN2 that were interesting is that it had a pretty bias, and that's because all the
photos of people that we had on Flickr tended to be
of attractive people. One of the biases that
AI models have today, and you've probably
noticed this, is that usually, it creates a person that looks
conventionally attractive. It's possible to work around these biases and in that way, raise them and raise the
concerns we have for them. As you're creating
images of faces with AI, it's important to remember the faces you're
creating are not real. If you want to add
authenticity to these images, you have to include
that in your prompt because it might not be as
present in the dataset. Another ethical consideration
that's important, and that we might
not solve today, is this idea of where
does this data come from. All the faces that are used to create new faces are pictures and images of faces
that we have all contributed to bringing
onto the Internet. They might not be collected from us in the most ethical way and sometimes those methods of collections infringe
on our privacy. These are all concerns
to keep in mind when you decide or if you do decide
to generate faces with AI. In my own practice, I tend to heavily stylize the faces that
I generate with AI. I make them more cartoonish
or more painterly because I want through aesthetics to bring my own touch
to these images. I'm not necessarily too
interested in creating images of people that would look real because that's something that makes me uncomfortable. Bringing a strong aesthetic into your process of making
images of faces with AI can be a way to work around
the discomfort you might have with creating portraits of people that would look real. For me, it works, and we're going to be looking at how
exactly I do that today. To start making this collage, I'm going to use a
bit of material. I have the images of
my family on deck. What I'm going to
do now is create some images to use
with Midjourney. One way to prompt is to
just use images as prompts. In Midjourney, it's
possible to do that. What I'm going to
do is I'm going to use an image of my grandma. In this image, she has
an awesome hairstyle. Then I'm going to mix that with a painting reference from what you might find in Senegal when you go to the barbershop. There's this really cool
trend of barbershops having these beautifully
painted portraits to show you different
hairstyles that they can do. I'd like that art style
and I'm going to use that with a real face to see
what the model gives me. I'm looking at the
results and some of them are a little problematic
and some of them I like. This one, for
example, leans more into photorealism,
which I don't want. This painting lacks definition, so I'm not really
getting the faces. I'm not getting the
hairstyles too well. This one shows a bit of a
racist trope, which happens. But when I've done
that previously, there was a pretty
interesting one that I might want to print out. Ordinarily, because
I like this image, I would keep going, exploit this image a bit more,
maybe play around with it, change the prompt, maybe add some more stylistic references
to get it to be strong. I'm going to use the
other way to prompt, which is to actually write
down what I want to happen. Here I'm going to prompt a middle-aged black
woman wearing glasses, sitting on a wooden stool with long braids that wrap
into a high bun. I'm going to add a
stylistic reference, in the style of West African
barbershop painted signs. In this way of
prompting using words, we get more of the model's aesthetic preferences
that come through. The images you'll see
in contrast to how we prompted using only images is that you have
more definition. It's a stronger aesthetic, but it's not necessarily one
that I'm more interested in. I actually like the images
that have a bit more chaos, that are more uncanny
valley like this. Thankfully, I've
done this before, so I've already printed out the images I'm going to
be using for the collage. I've printed two sets of images. The first set is
pictures of my family. This is my mom, this
is my grandmother, and this is an ancestor
or potential ancestor. Then this is images that
I've created using AI. Now we're going to make
a collage with them. I'm going to cut out
parts of the image that I think are interesting. At my grandma's feet, you have a chicken
that I quite like. We're going to take this part of the image and set it aside. We're going to take
this part of her face, the one where we can
see her glasses. Another cool thing about
the way these images are printed is that they are
printed at different scales, so I get to play around with making
these portraits quite abstract and absurd. Now we have this neat
little face that's missing eyes and we're
going to attempt to replace her eyes maybe with
some AI-generated eyes. Let's see. I'm going to take a colored paper to
start this image. I don't glue them until I have
some composition in mind. I layer them and I get to move things around until I'm done. Now we get to try this out on some of these AI-generated eyes.
I quite like this one. Let's try it with the last one. There's a bit of attitude here, but they're not
centered on the face. I like that. This is our one. I am looking at the
different images, and I think I quite like
these earrings here. Let's maybe glam her up. I quite like the
blue of her cloth, and so what I'm going
to do is quickly sketch out an outfit for this and
then cut it out of this cloth. Normally, I would
keep working on this, but you're getting the picture. A person is starting to form. We're going to add our
little ear and earring, and I'm going to keep
working on this. Using the same technique, images of people you know and images that you've created, try to come up with some
collages and try to create some portraits that are fun and interesting with
this technique. Meet me in the next lesson
where we're going to talk about creating
environments using AI. [MUSIC]
6. Building Environments with AI: In this lesson, we're
going to be talking about building
worlds with AI and why it's important
and interesting to bring in speculation and
fiction into your practice. We're going to be
generating a 3D asset in Nomad and then we're
going to apply some textures to that
3D assets and bring it into a world that we will
have imagined with AI. So to start, we're going to open Nomad and I've already started
sculpting this 3D asset. In this first part, it's
all me, it's my hand. I'm creating something
from my imagination. Then eventually we're going
to bring in AI into it. But to start, we're taking
a dive into my mind. I wanted to create a speculative
monster that I was going to add into my very scary world. Like a horror phase, I guess. So this tool allows me to sculpt in a pretty intuitive
way 3D assets. You can obviously use
any other tool that you used to making 3D assets, but I'm going to use
this one called Nomad. So first, we're going to maybe
add a bit of a dent here, make the tentacles
slightly larger, then we can smooth that over. I don't have a crazy, clear idea of where I'm going, but I know that this is somewhere between
a mask and a monster. It's a little bit inspired by masks you would find
in West Africa, so I'm using a more
metallic material here because I want to get these play of light
that you see here. Obviously, you can dial those
up or down in this program. But we have good starting point. Now that we have our assets, what I'm going to do is, I want to make an
art piece with it. So I'm going to render it as an image with a
transparent background, and then I'm going to save that into my laptop and
then we're going to insert this new artifact that we made into an imaginary world. So when I look at my artifact
or my little masks monster, I want to find a home for it, and so I need to
make that world. Which world does it live in? I'm going to take
some images that I've previously created and then I'm going to try to create a
cityscape using those. To help me do that, I'm going
to select an actual image. This is a satellite
image of my city and then I'm going to blend that
with this image reference, which would be here. Then in the prompt section,
because I can do both, I'm going to add a
fantastical city where nature is deteriorurating, dark color palette, abstract, shapes that reminisce
of decaying nature. Normally I would do
this process over and over again until I get
an image I'm happy with, but we're going to go with whatever we get
the first time around. Let's see where we got. We
don't want to use this image, a little too cartoonish.
I like that one. That actually blends in
the reference really well. I like the decaying
building in the image. Not so crazy about this one. This one is a little bit too abstract. We're going
to use this image. So let's save that
and then we're going to import both of our images, both the assets and
the world we just created into our iPad. Open Procreate and make sure
that fits into our Canvas. Good. We might want to play
with the colors a little bit, maybe add some texture, let's do that. Happy with that. Maybe play with the
curves a bit more to darken it while still improve. I like low contrast images. So I'm going to do that. I think it helps make the
image a bit more abstract. I'm happy with that, so I'm
going to add the asset now, which should be here. So now we want to add some
texture to this new asset. So I want to create a mask that allows me to make it
a bit more interesting. Let's play around
with that a bit. I like these textures. It's basically a texture
that I've generated using AI in the colors that I
would like for this image. Then I'm basically going to
use that to give it a skin. We can think about it as a skin, play around with it a bit more. So I'm happy with the result. The texture is very faint. I'm basically playing
with brightness and saturation at the moment to give it the hues that I want. Then maybe I'm going to add some noise as well in order
to harmonize the two images. The next thing I want
to add to my mask is maybe give it more defined
features like eyes. I'm going to still
stay in a very abstract by blurring these out, for example, we'll
continue working on this, but this is basically
the process. I have a 3D asset. I have a world that I
want to bring it into. I'm going to play with the
placement a little bit. Maybe I'm going to create more of it because right
now we have the head. Now that I see it in place, I wanted to have a body, so I'm going to create that
and then bring it into it. Maybe I'm going to add
other features using Nomad. This world is starting
to form together. I have the character. I have the environment, I can keep adding to it. I can keep expanding the image. I could also use this if I want inspiration
for making a game. But this is basically
one way to start generating ideas for 3D assets and for world building using AI. Whether I'm creating
the 3D assets or I'm selecting the images that will eventually create the
world that I'm making, my taste really comes into this, my influences, what I'm
interested in showing, and it's really about
leaning into what you like and getting there using some of the tools
at your disposal. The other thing I
love about doing world building in
this way is that I get to step into speculation and fiction in a way that I wasn't able to
necessarily before. I, for example, can't create a
game from scratch, but I can start to create the building blocks of that
game using this technique. In general, I think speculation
makes me a better artist. Here's what I want you
to do. I want you to think about a world that
you've been dreaming up. Maybe that's an inner
world or an outer world, and it's come up a few
times in your life and you've never really
tried to bring it to life. You can try and bring
that together in whatever software you
feel comfortable with, whether that's Procreator
or illustrator, or whatever software makes
you feel comfortable, or even if you wanted to create a collage using some
of these assets. Meet me in the next lesson, where we're going
to be talking about reimagining in the past with AI
7. Reimagining the Past with AI : In this lesson, I'm
going to be training a small personalized AI
model with images of what I think has been
missing in terms of image references when I'm working with AI,
which is braids. They're very common
where I'm from, and I wanted to go
back in history and see if there
are archival images that seed some of
the styles that we see today and use that
to create a model. The results of this
model will be used in the scene we're going to be making at the
end of this class. For this lesson,
we're going to be taking a look at public
domain archives. We're going to specifically
focus on Spawning, which is one of my favorite, and we're also going
to be training a LoRA using Civitai.
Let's get started. First, we're going
to be looking at a public domain archive of very high quality
images for training. This is specific if
you're an artist and you need really good
images to train with. Some of these have
really high resolution. It's called Source.Plus and it's one of Spawning's products. As you can see, there's loads of images and you can
search by keywords. For example, I'm going to search the country that
I'm from, Senegal, and there are
hundreds of images, some of nature, some of the architecture,
some of the people. Depending on the model
I want to train, I can find something very
specific to my need. There's another cool feature which you can search by image. I can upload an image. Let's use one of
the images that I had earlier and see
what we might get. If you have an image and
you want to trace it, and you want to find
related images, you can also use that feature. The computer vision
is pretty good. You can see that it's recognizing that we're
in the art realm, we're looking at a painting, and so it's giving me things
that are art related. That's one of the ways
archival research works. You're either using text
reference or you're using image reference to look for images that you can
use for training. There are loads of sources
for public domain images, but this is personally
my favorite because of the high resolutions
that the images can have. To train my LoRA,
I'm going to need some images of illustrations
from the past, looking specifically
at illustrations related to the image that I uploaded of the
barbershop paintings. What I'm trying to get at, this is experimental, is whether there's
a common aesthetic to these paintings that I can investigate and try to
replicate with the images. At first, I'm going to select
a few images from this. We're going to go with 10, and I'm going to focus
specifically on hairstyles, and then we're going to use
that to train the LoRA. First, I'm going to add
images to this collection, so we're going to call it
Historical Hair References. We're going to save that, and we're going to add
it to the collection. We're going to focus
on hair references. I'm going to click on images
of hairstyles that I find interesting and I'm
going to add them to this new collection called
Historical Hair References. I'm going to keep doing
that. That's a good one. This was possible because I uploaded the image of the
barbershop hairstyle, and it was able to do a reverse research and
show me related images. I'm getting a lot of
really good references that are all public
domain and high quality, which is the best thing
you want for training. Some of the images
in this dataset are very interesting
to me and some of them offend me because
they were taken by perhaps someone who's not
familiar with this context. We now have 12 images that we are going to download
and use to train this LoRA. Let's export that. We're going to caption our
image historicalhairref. I'm going to export
that into a zip. Now that I have my images
downloaded locally, what I'm going to do
now is going to use it to train a LoRA. I'm going to use the platform Civitai in order to do this. You can sign up for
free and you can train, I think, a very basic LoRA. But if you want to train it faster or add more credits,
that's going to be up to you. But there's a basic free
version for this tool. We're going to call our LoRA
Historical Hair References. What this LoRA is,
it's basically a very small personalized model built on top of an
existing larger model. What it will do is learn the aesthetic from the
set of images that I have, and then I'm going
to be able to use that dismal model to prompt for more images based on the aesthetic and the patterns
present in the training data. We have an option
between creating a LoRA that makes characters, that makes a style, or that makes a concept. This is a tricky one because I'm focused on the
hairstyle specifically. It's not necessarily
the characters, I'm not so much interested
in the face of the person, I'm interested in the
object of their hair. I'm also interested in the
style that these images have. It's going to be between style and
concept. I can do both. I can start with a style, see the results I get, and then train
later with concept, and then see which
result is better. I'm going to start with style, and then add my training data, which will be right here. Now that we have
uploaded our images and we can see them down here, what we're going to do
next is label them. We're going to look
into the folders from our downloads and the labels
should all be in text files. We're going to copy that over, and keep doing that for
the rest of the images. I like Source.Plus because it gives me the captions
of the images as well, which is really
important because when I'm going to be
prompting this model, the model is going to
rely on the words that are present as well
as the images, and so it's going
to make some of the results a bit
more accurate for me. No archive is perfect, and we're going to
look at a really good example of what I mean. When we look at this image, we know for a fact that this is a hairstyle and that
this is done with hair. But when we look at the caption that's generated by
computer vision, we see that, it says this image shows a black and white photo of a woman with a large
hat over her head. This is a clear example of
how the model is going to misinterpret this image and
label this as a hat when it, in fact, an actual hairstyle. What we're going to
do is correct it, which we get to do when
we train our own models. We get to make corrections based on what we know of the context. Let's do a reverse image search, and then we're going to
locate what's happening here. It seems to be a
Mangbetu hairstyle. We're going to take a
description that we see here and then we're going to use that
as a caption for our image, funnel shaped coiffure that
ends in an outward halo, Mangbetu hairstyle, woman's face is in focus while
background is blurred, woman is wearing earrings
and a beaded necklace. Now we feel better
about the caption. Now that we have
selected the LoRA type, we have added our training data, we've added our caption, we've seen how to make
the captions work for us. What we're going to do now is choose a base model
for training. For the purposes of this video, we're going to leave it at
the standard SDXL model, but you can also take
your time and look at all these different
standard diffusion models or use a Flux, which is one of the newer ones, and then see which one gives you the best results for what
you're trying to do. This is going to take
about 45 minutes to train, so we're going to
leave that for now, and we're going to see the results together
in a different lesson. We've seen that there are
two ways that you can approach thinking about the past and using AI to help you. You can either curate your own datasets with
images that you've taken yourself or maybe
archival images of your family, things you have lying
around the house, or you could explore
historical archives, public domain archives out
there on the Internet. Once you have those images, what I would love for you to do is to go ahead and
train your own model. Using the Civitai platform or any other platform
you know where you can train your own LoRA, it's a really
interesting exercise for you to have to deal with, one, gathering this data,
labeling it properly, making sure you feel comfortable about how you've
acquired this data, and then using that to train. One of the really cool
applications of this type of work that I've seen is for us to have an opportunity to tell stories we haven't heard being told enough or we've never
seen being brought to life. For instance, I used it to
reimagine what flowers that my grandmother grew up with
might have looked like because they're extinct and I have no images of them today. You could do the same.
Think about histories that would empower you and how you might be
able to retell them. This is something we're
going to be discussing and including in the final scene we'll be making at
the end of the class. If this is your first
time making a model, I'm really excited for you. Go ahead, do it, have fun, and meet me in the
next lesson where we're going to be
imagining futures with AI. [MUSIC]
8. Imagining Futures with AI : In this lesson, we're going to be taking ourselves
into the future. Literally, we're going to
shoot a video of ourselves, and turn ourselves into
a future character. We're going to be including
some of the shots from the video we're going to be producing into our final zine. In my version, I'm continuing
the theme that I started in the previous lesson of looking at hair as
a reference point. I'm projecting that into
the future and creating a character whose hair is
channeling information, which is something braids
have done over the centuries. During the time of
slavery and colonization, women all across West Africa used their hair to
braid several things. They would braid seeds
into it that they would transport on their
voyage across, or they would also
use it to map out all the exit routes and escape the plantations
they were kept in. This is an interesting
concept for me because we can think about hair as cables, or as information processing or information
channeling devices. I'm quite literally
exploiting that into an image of a person from the future
whose hair is bionic, essentially part organic
and part mechanical, and can conduct information. When we think about the future, we often think of a very specific idea that's been probably
fed to us from sci-fi. What I would encourage
you to do is to try to think about a future
that feels right for you. What are you
interested in seeing? Is it more organic? Is it more synthetic? Are there things that you want to bring back
from the past? Are there certain things
that you're missing in your life right now
that you'd like to see? For me, it had to do with hair, and I really wanted to play
around with that because I like to do my hair in
different ways every time. But it could mean something
different for you. Really think about
what kind of future is this character
in and really try to drill down on specific
objects or decorations, things they're holding
on their bodies, things that are part
of their bodies. Really try to think about what their embodiment looks like and how it shows which
era they're in, and helps them in
that era in some way. Go ahead, find inspiration
and go wild because, honestly, the only thing
stopping you is how far you want to take your
imagination on this one. We'll be using Runway specifically
Act-One's feature on Runway to create our
character video today. To begin, you can create an image of a character
that you want to be using. I've created mine
in the Journey, and I will port that over
into Runway to get started. Here's Runway. I have a Pro's subscriptions, but there's a free version, so you can use that, try to see if you want to get
the subscription yourself. What we're going to
be doing today is use this feature called Act-One. Let's try it now. There's a couple of
things that we need here. First, we're going to record a video of ourselves
talking about something. Then we're going to import
the image of a character. There are some character
suggestions here that you can use if you
want to test it out, but I'm going to upload my own. I'm going to now
upload an image of a bionic woman that I've
created in the Journey, and this is what she looks like. I'm going to crop that out
to make sure we can see her hair well. We
have our image. Next, I'm going to record
myself saying something, and we're going to see the
magic happen after that. A few things to
note here is that my character is facing
this direction, and so it's a profile shot. I'm going to need
to turn my head slightly in order to give this model the best chance
to create something. I'm going to record
something and I'm going to say, "Hello. Hi, my name is Linda, and I'm from the future." Here we have it. I'm going to move it
back a little bit. I'm going to step back
so my face is in frame. I'm going to turn
my head slightly. Hi, my name is Linda, and I'm from the future. "Hi, my name is Linda, and I'm from the future."
Go ahead and try it out. You can generate videos
up to five seconds. You can string them
along. You can create longer videos
if you want to. Ours is done, so we're going to take
a look at it together. "Hi, my name is Linda, and I'm from the future." There's a couple of
things that I like and a couple of things that
I would want to change. One of them, I think, is the angle that I used when
I was shooting my video. I think it could have
been slightly more aligned with the
photo that I have. But the result is still there. There's something to exploit here. The character is moving. I can imagine making images where there's a bit more of the background that's showing a bit more of the environment. Yeah, this is a very
interesting first experiment. I'm going to keep playing
around with this, and then take some
screenshots of the moments that I really enjoy, and reflect on what this experience meant for
me in the final zine. Meet me in the next
lesson where we're going to be assembling our
final zine together. This might include some of the images from the
hair exploration, or some of the screenshots from the video we just
made in this lesson. We'll get to look at
everything and then choose what makes the final
cut. I'll see you there. [MUSIC]
9. Creating Your AI Manifesto : In this final lesson, we're going to be assembling
our zine together. This is going to
be our manifesto for how we think about AI, how we plan to integrate it
in our practice, if at all, the concerns we
still have for it, and the hopes we have as well. You're going to need
whatever program you're going to be using
to assemble your zine. For me, it's going to be
Illustrator because I'm assembling it on my computer
using digital files. But if you prefer to use physical files and you want
to make a physical zine, you can print out all
the assets you've made, and use that to
assemble your zine. Before we start,
please bring next to you any journal, entries, notes you've made
along the class, especially our five-part
diagram about our motivation, values, hopes, concerns,
and questions. In my case, I have
both of my journals. In this journal, I will have all the questions
and all the readings that I've done on AI that I've recorded over
the past few years. I also have this
newer journal where I've been recording my
thoughts along this class. Gather everything you need and let's get started
assembling our zine. I've titled my zine
Making Memory Machines. These are three words
that shuffled in different order speak to me about how I use
AI in my practice. My zine is going to
have about 12 pages and I have my cover image
and my back cover. I'm going to get into the
first couple of pages. This first one is
titled Saudade, and it's basically the
Portuguese word for nostalgia. It's part longing,
it's part hopeful, and it's a complicated
word that I think for me signifies one of the
most important value that I bring into
my work with AI, which is making sure my past, which to me represents myself, is brought into my practice
when I work with AI. We're going to be using
one of the collages we made at the beginning of
the class into this page. Then we're going to look at
notes that are related to it, maybe find some quotes to
add into this section. There goes my first image. I'm going to size
it appropriately, place it on my page.
I think that's good. If you remember from
an earlier lesson, I started making a
collage physically, and I've continued that
collage digitally, and that's what we have here. It includes plants from
our garden growing up. It includes broken bits of
the floor tiling in my house, which was this black
and white checkerboard. It includes a composite
face that, I think, looks a lot like
women in my family, but lacks detail, has painterly feature that I worked on and
stylized a bit more. That's the image that's
going to be here to represent this idea of Saudade. Then in my notebook, I'm going to find
decent quotes that I want to bring into
it. I like this one. This section talks
about my perception of time when I make work
with technology, especially given the pace of AI, and so I'm going
to include that. I guess my exploration of the past is about hanging
on to this timelessness, and telling myself
that I still have time to explore who I am by
diving into the past, and I don't have to
rush into the future. Now I'm going to find
a couple of quotes. I like this one. This one says, "Computers that think for you. If you believe they
are ultimately simple, you shouldn't endorse
such answers." I like that. I don't want the computer to be
thinking for me. Now I'm going to write down some of my thoughts on what it means to me
to create faces with AI, and how that relates
to my idea of Saudade. "There is still time to learn to make
different choices in how AI evolves and to develop our agency for
how that happens." Then finally, "Computers are
only deputies for thought, they are not thought itself. I don't want computers
to think for me. I will not endorse
a future where they do my intellectual
and heart labor." In this new page,
I'm going to be reflecting on how
AI makes images, and what that means to me, and to understand
the mechanics of it means to empower myself to think of the way
I produce images as a completely
different process. But these two processes don't
have to necessarily clash, they can exist together
into one practice. That's what I'm going to
be reflecting on here. For this one, I'm going to be adding a conceptual artwork that has to deal with this
denoising process that AI uses to create an image. I have these blurred out portraits that
I'm going to include. I'm going to arrange
them in a grid. Starting with the
more pixelated image at the end or the
more noisy image, and then working
backwards from there. Now that I have
this section ready, I am going to pop
that over here, and then I'm going to
look for the words. We're going to keep
doing this over and over until our zine is complete. I like this one by Wendy Chun. It says that software is a cultural practice
made up of algorithm, human interaction, and
speculative imagination. The reason why I like this
quote a lot is because we often just think of
tech as these hard, cold algorithms that
aren't ruled by anything, but they are made by humans
who have certain values, who have certain ways of thinking about the
world that guide them. Then there's this
Weizenbaum phrase called reckless
computational imperative, which is the very masculine
culture driving computing which pushes towards evolving
computing at all costs, ecological costs, cultural
cost, human cost. I think that's something I
definitely want to talk about. Let's write down
one phrase here, and continue to
explore it later on. "Reckless computational
imperative. What does it mean? Who makes these decisions? Who is affected by
the recklessness?" Let's put that over here, and then continue to
work on the next page. In this page, we're going to be talking about what does
it mean to build world? What does it mean to
think about the future? Speculating is my right. Because this is
about speculating and thinking about the future, I'm going to be adding both
the artworks I've made for the environments lesson as well as the 'imagining
the future' lesson. I'm going to first
set the setting here with one of the
environments I created. We want this to be really
prominent on the page. Then we're going to add a little screenshot of our future character
that we created, and I've kept a little bit
of the audio player here to remind me that this
is an animated piece. I'm going to collage
that in right there. Then because this is the most important
message of this page, we're going to make
it more prominent, and get it to take up a
lot more of the page. Almost like we're
making a poster. Then we're going to
add another phrase to reclaim the future. "The future is
mine to dream up." This page is going to be very important when I
lose motivation, to keep bringing
myself into the work, and so I'm hoping it's
a North Star for me. Now we move on to
looking at the past, and reimagining the past. For this one, we get to look at the fun results we got
from training our LoRA. These are all images that the AI has produced from
the model we trained. If you remember in the
previous lesson with the 12 images we used
to train the LoRA. I'm not terribly
happy with it because while some of the
training images do relate to the
reference images, the model created new images
that are overwhelmingly skewed towards a Caucasian
female embodiment, which wasn't present in
the training data at all. We're going to mark evidence
of what we've gotten, and use that in our
zine to remind us that models are biased. I'm going to be adding one of the reference images that
started our whole journey, and then I'm going to add
where the model got us. We're going to write a
little descriptor here. We're going to divide
the page into; and the theme of this page is going to be about
retelling the past. Here we want to get
across this idea that datasets aren't created in an apolitical acultural vacuum. Then we're going to refer to
a quote we had earlier about technology being made
of human interaction, algorithms, and
speculative imagination. To remind ourselves of
where the biases come from, and why we should strive to
always be aware of them, and work around them,
and denounce them. We're going to have that here. Then we're going to describe our LoRA experience with this model. "I trained a LoRA using historical references
of hairstyle from West and Central Africa. However, the model created mostly Caucasian
representations. To get to a future where
this doesn't happen, I have to continue to increase my agency in how
I work with AI." I'm going to keep
working on this, and I'm going to
keep adding images. The last section of this manifesto is going
to have some principles, five principles that I have, that will continue to guide
me as I work with AI. I'm also going to jot down some of the questions
that I still have and opportunities
for further exploration, and those will be
my last two pages. The five principles that
guide me while I work with AI that are going to be present in my manifesto
are the following. "What is tech doing to
me or to people like me? Bodies, minds,
relationship, dreams, plans, way of life, history,
psyche, philosophy? Investigating that is very important to me and it's
central to my practice." "While my work may
have some activism, I am an artist,
first and foremost. I ask questions, I reflect what I learn and
unlearn in my work." "I still have time.
There's always time to learn and to make different choices and
to develop my agency." "I strive to bring
art where I'm from. I take from where I'm from, and I bring it back
to where I'm from." Then the last one is, "Ultimately, art
is about feeling. If I can make other
people feel the way I do, that's enough for me." Then on the last page, I would write some of the questions that I
still want to explore. For example, "What
is the ideology or the ideologies that guide the way tech is currently
being developed?" "Why don't we talk
about all that AI is, including the material
reality of it?" "Is there an
ecological future for our planet given the current status quo
when it comes to AI?" Those are just a few examples, add more questions
as you go along, write out everything you
want to keep exploring. Now it's time to go and
make your own zine. Write out your manifesto for how you want to use AI
into your practice. Or maybe you finish
this class and you decided you never
want to use AI again, write a manifesto for that also. Hopefully, your zine
is a document that continues to live and
continues to guide you, and you should keep
asking yourself these questions even
after you've made it. Please upload your final
manifesto zine into the class gallery so that you can inspire other
students to do the same, but also draw
inspiration from them. I can't wait to see
what you've made. Thank you for taking this class. [MUSIC]
10. Final Thoughts: Congratulations. You've made it to the end of the
class. Well done. We've made people faces. We've built environments. We have imagined the future. We have brought
ourselves into the past. In the last few lessons, we've explored a bunch
of different AI tools, and thank you for creating
your AI manifesto zine. I hope it was a
fruitful exercise for you that you
got to reflect on why you want to integrate AI
into your practice or not, what questions you
still have about it, what your hopes and fears are. I hope you keep asking yourself those questions and referring to this document and perhaps the documents of other
students as well. AI is scary, but hopefully, over the last few lessons, we've made it a little
bit less scary. When I started working with AI, I did so as a way to see myself
represented in it and got really curious about it and very excited about the
tinkering potential it had. That doesn't mean I'm okay
with everything that it is. I'm constantly uncomfortable. I'm constantly asking questions. I'm constantly doing more
research in order to feel like I'm
developing some agency. As an artist,
especially an artist from a marginalized community, it's so important for
you to keep showing up and for us to keep bringing
ourselves into this tech. Because of the way it evolves, because of the way
it's distributed, if it doesn't include us, we
have to include ourselves, and we have to make a stance, and we have to show
them what it means when tech represents a diverse
and plural world. That's exactly what we've done
over the past few lessons. You probably will figure out
ways to do it yourself in your practice as well as you
continue to work with AI. Hopefully, now you're
walking away with a little less fear about
what AI can and cannot do. You're starting to look at it as just another tool in your studio environment that you can bring into
your practice. It doesn't have to
be intimidating if you continue to inform
yourself about it, if you continue
to read about it. Whether we like it or not, AI isn't going anywhere. It's just going
to keep evolving, so it's up to us to determine
how we want that to happen. If you're going to
use your work to critique it, please do so. If you're going to
use your work to show us a better version of it or a better way artists can work with
it, please do that. It's important for us
as artists to bring our voices into this
conversation and start talking about it to a
wider community because so often we're the ones who reflect
culture back to society. I think it's a privilege
to be able to do that, and AI is a big defining
technology of our time, so let's talk about it. Please dream up a
different future for AI and do that
through your work. Please don't forget to upload your projects
to the gallery. Thanks for taking this class. I cannot wait to see
what you've made.
Linda Dounia Rebeiz, Artist, Designer, Writer
