AI for Designers: How It Works, Prompt Writing & Design Inspiration

1. Course Introduction: A lot has changed in the AI space over the last few years. When it first came out, most of us, myself included, saw it as a novelty. Fast forward to now, the AI models have evolved to the point that it can almost replicate a human creative or at least temporarily fool them. Almost every large tech company has invested billions into the AI space, and that has led to better AI models, training and research. There is this fear that AI will take over a creative person's job. That fear is very real, and photographers, graphic designers and illustrators are allowed to be a bit worried. I decided to create an AI theory course that alleviates that fear. I spent hundreds of hours utilizing the latest AI tools and models, and I come away very hopeful that AI will be a wonderful complement paired together with our own creative skill sets. This class is a good starting point for any creative or designer who wants to learn about AI basics, like the behind the scenes of how multiple AI models work, like image generation tools and chat AI models, prompt writing basics and how to write effective prompts, learning nuanced design terms that provide us with the right vocabulary to craft hyper detailed imagery. We will discuss the legal issues, too, and more. Learning to write prompts is like learning to write all over again as a child. We start with basic prompts that lightly describe what we want, and eventually through practice, we're able to properly describe in immense detail what our vision is using words we may have never thought to use. This class can appeal to almost anyone who has been intimidated or overwhelmed by AI and did not know where to start learning the basics. This class is designed specifically for creatives and designers to help guide you through AI basics, so you can start to think about how to utilize these amazing tools to help align yourself with the future evolutions in the creative industry. My hope is this class will be a springboard for future classes that will go over practical design projects using AI. But first, we must understand the basics behind how it works and this class aims to do just that. I will see you in the first lesson. My name is Lindsey Marsh, and teaching design theory is my jam. I've been a graphic designer for over 20 years and a design instructor to over 350,000 graphic design students. I'm excited to be able to bring this class to you today. 2. How LLMs Work: Let's generate a red cardinal on a branch. How did this image generator know to show a bird? How did it know the bird was red and was on a branch? How did it even know what a branch looks like? This is a complex system that runs millions upon millions of precise matrix calculations to produce what seems like magic. To really get a good idea of how this happens, we have to first understand how AI models understand human language, to then understand how it will generate imagery. So what is an LLM? An LLM or a large language model is a type of artificial intelligence trained to understand and generate human language. Tools like chat GPT, Claude, and Google Gemini are examples of LLMs, are large language models. I'll be using chat GPT throughout the course, which is also one of the most popular LLMs out there. They're called large because they're trained on massive amounts of text, everything from books, articles and websites to online conversations and more. The goal to learn how humans communicate, our sentence structure, grammar, tone, style, and even intent. So the AI can respond in a way that feels natural and useful. Sue, why do LLMs matter to designers like us and other creatives? Because LLMs are more than just writing assistants. Here's a few ways in how they can help or design workflows. First of all, idea generation. Need a concept, a slogan or a campaign direction. Just ask. Copywriting. LLMs can help draft social media post, taglines, product descriptions, and more. Creative briefs. You can get help structuring or editing client facing documents. Naming generate brand names, product names, project titles, everything based on tone and keywords. But more than just that, learning about how LLMs work allows us to understand the all important tool for designers in the next decade. The image and video generation tools we'll be using throughout the course. So in the course, we'll focus on two types of AI models. The first one is the one we just mentioned, the Large language model or LLM, and that's eventually fed into another system, which is called the image diffusion model. And the image diffusion model is what helps us generate images from text. For example, it can recognize that Apple relates to fruit without being explicitly taught that connection. This ability to interpret language is essential for image generation tools, which we'll explore throughout the course. Before an AI can create a visual of a dog barking, it must first understand what a dog is and what it means to bark. So the large language model learns and understands and generates human language by analyzing massive amounts of text and finding connection points between them. And then it's fed prompts into an image to fusion model which generates images guided by patterns that learn during training. It often relies on an LLM to first understand and interpret the text prompt, converting it into meaningful tokens that will guide the visual generation process. Now let's talk about how LLMs work. Think of it like a supercharged autocomplete that does not only finish your sentences, but can write essays, answer questions, design prompts, and even help with branding and copywriting. At their core, LLMs are probability machines. When you ask a question, they calculate which words are most likely to come next based on everything they've learned. For example, it's like a seasoned designer who's so used to trends, client needs, and layouts, they can almost guess what the client wants next, even before the client even tells them, because they've done it over and over and over again. The next is generating tokens and context. So LLMs don't see a whole sentence. They break them into little chunks called tokens. Words, part of the words, or even punctuation matters. Even the purity at the end is its individual token. They then look at the context, the text around it to figure out what's likely to come next. And then next, there's several layers and processes it's run through. LLMs have millions or even billions of neurons, mathematical units that process language in layers. Each layer refines the understanding of meaning, just like a creative review process. So let's go through that process in more detail. So let's have an example prompt. And this example is create an image of a furry dog. So it's good to divide each word into tokens. So create would be its own separate token and image of a barking dog. Occasionally, it'll divide a word. So barking could be bark and then, and also periods count as tokens as well. So each word or token is given a vector point. LLMs don't understand words the way humans do. Instead, they represent words as vectors, which are like long lists of numbers, sometimes 12,000 long. These numbers capture the position of a word in a massive invisible space called the embedding space or vector space. Each word becomes a point in the space where similar words are placed close together. In our prompt example, the word dog and barking would be close together in this map because they were frequently shown associated together during the training by the data. This example uses a two D space, but AI models have 50,000 words to map out. So there's not a lot of room. So what it does is it maps everything out in a three D vector space. This is why each token or word is assigned a long list of numbers as these pinpoint the exact location on a three D vector map. These columns of numbers are coordinates that allows words to find each other and therefore develop associations and human language with each other. Imagine a giant three D cloud, except it's actually thousands of dimensions large in this space. Similar meanings equal closer together. So King is close to Queen, and Paris is close to France. And designer is near other words like creative and visual and artistic. Different meanings are farther apart. So King is far from Apple because King and Apple don't appear together very much in human text and language. Light brightness is a different area from light weight, depending on context. This map of meaning is built during training as the model learns how words appear in context. LLMs don't understand words in isolation. Instead, they consider the tokens around them. So for example, the designer used light colors in the layout. Here, light is interpreted as brightness because of nearby tokens, color, and layout. But the backpack is very light and easy to carry. Now, light means not heavy thanks to the context words like backpack and carry. The model dynamically adjusts understanding based on context, and it does this through a mechanism called attention. So let's talk about attention. Unlike older models that process each word independently, attention, which is part of a bigger transform layer, which we'll get to is another process that is run that lets the model look at all the other words in the sentence and ask, which of these should I pay attention to in order to understand what this word means? It gives the model the ability to weigh words differently depending on their relevance to the word it's generating or analyzing. It's like a designer reviewing an entire mood board before making a decision about a single layout element. Because meaning often depends on context. For example, the word bank can mean very different things. She sat by the bank at the river. The attention function highlights the word river. He made a deposit at the bank. Attention now highlights the word deposit. The model uses attention to focus on the words that clarify which meaning is correct. And then the data after attention, flows through lots of other different processing layers. And a lot of these are different mathematical matrices calculations that you're seeing all behind the scenes that are happening millions and millions of times. So the next thing is the feedforward neural network. So after attention, each token's updated vector on that little three D map I showed you now is enriched with more context. It's passing through a small neural network called the feedforward layer. This network applies a mathematical transformation to the vector. It doesn't mix tokens with each other. Each token is processed independently here. Think of it as a refinement step that helps distill more meaningful patterns from the attendant information. It's polishing it up, it's adjusting and fine tuning and enhancing it before it's passed on. I wanted to take a moment and pause. This is a very complicated mathematical process with lots of layers that are processing data over and over. You don't have to be a mathematician to understand how they work. I just wanted to show you a little bit of a detailed guide of how they go through the processes, but you in no means need to memorize this or know this front and back. It just helps us later on when we write prompts to know what's really going on behind the scenes with how it's processing our words. The next step is residual connections, skip connections. This is to make sure the model doesn't forget the original information. It uses residual connections. These are like little shortcuts that add the original input vector back into the output of each layer. It prevents the model from overwriting useful information with too many transformations. So it's kind of like editing a design, but keeping the original version as a backup layer in Photoshop. Once again, you don't need to know the mathematics behind all this, know that this is a very complex process that happens and why AI sometimes seems like magic. There's lots of checks and balances that happen to make sure what it's coming out with is checked and reviewed. Another layer is called layer normalization, and this is a cleanup step. It helps stabilize the training and keeps the data consistent across layers. It ensures the model doesn't get too biased with extreme values. For example, it's like adjusting levels on a photo to even out the lighting before moving on to the next edit. So we're stacking more and more layers of processing. And transformers don't just do this at once. They repeat this whole process multiple times 12, 24, or even 96 times depending on the model size. It's going to go through the attention. It's going to go through the feed for neural network, the residual connections, the layer normalization, and it repeats over and over. So each layer builds a more nuanced understanding of human language. So lower layers, understanding structure like grammar and punctuation, the middle layers, recognizing meaning and relationship, and some of those higher layers of processing that are later on help with reasoning, planning and completing tasks. So, for example, it's like going from sketch to refined illustration to full brand identity. And we have a final output. After going through all the layers, the final vector is used to predict the next token for text generation, token meaning word, classify something like a sentiment or a topic, or guide image generation, like in a diffusion model, we'll talk about next. We just scratch the surface of how LL models work. But if you really want to get way more technical and dive into the mathematics, of course, not required at all for this course. You can check out three blue one brown on YouTube. This is how I first learned the details of AI models, and I found him a really, really good teacher. So we saw this complex weave of processing. The vectors representing words pass through many layers of data processing. Eventually, they reach a probability matrix where the model determines which word is most likely to come out next. While the underlying math is complex, what you really need to understand is just how deeply layered this process is. Each word is broken down, analyzed, cross checked against each other through multiple internal checks and balances. The result feels almost magical like the machine truly understands and interprets human language. This same kind of layered intelligence is what powers image generation as well through a process called diffusion, which we'll explore in the next lesson. Make sure to download the PDF resource that goes over everything we talked about in this lesson. 3. How Image Diffusion Works: AI Image and video generators have wowed the Internet for the last few years with their ability to fuse together objects, subject matters, and challenge social norms. The negativity surrounding these AI tools is slowly dissipating as creatives start to realize how critical they may be to keep up with the shifting and changing industry. Today, we get to learn how image generators work step by step, so we can see the magic behind the curtain. So how do these AI image generation tools work? So from language to images, how understanding LLMs helps you learn diffusion models. So in the last lesson, we spent a good 12 minutes learning about LLMs and how they process tokens. They build context, they apply attention, and they generate predictions. So learning that, you've already laid the foundation for understanding how diffusion models work. While LLMs generate words, diffusion models generate images, and they rely on many of the same core ideas like layered processing, high dimensional vector spaces, token like representations, and probability based outputs. The key connection is prediction as a core mechanism. At the heart of both LLMs and diffusion models is a simple yet powerful idea. Learning to predict something based on context. I LLMs, the model predicts the next word. In diffusion models, the model learns to predict a cleaner version of the image step by step from noise to clarity. Both systems refine guesses based on what they've learned from massive datasets. One works in the language space, the other in pixel space. So step one, the training phase, it learns by destroying images. So let's take a real example of a picture of a cat. Let's add random noise to it little by little, over hundreds of steps. As a designer, you may find this process familiar looking because it's the Gaussian blur and grain filters in Photoshop. Eventually, it becomes static, like TV fuzz. The model learns how each step of noise affects the image. This is thousands of different tiny little pixel additions and removals. This teaches the model how images fall apart. Step two, it learns how to undo the noise or the grain. The model is trained to predict the clean image from a noisy one. At each step, it guesses. If this is what the noisy version looks like, what might the original have been? It gets better by comparing its guesses to real images and adjusting its weights. So imagine a sculptor standing before a large block of stone. At first, it's just noise, a solid chaotic mass with no recognizable form, a big block of stone. The sculptor begins chiseling away little by little, guided by intuition, training, and references. Each strike removes uncertainty. Slowly, shapes begin to emerge, a curve here, a silhouette there. Over time, the noise becomes form and form the randomness of the block. A detailed and beautiful statue is revealed. This is called reverse diffusion. You start with noise, you refine it into a coherent image one step at a time. Step three, generation begins. It's the sampling phase. Now that it's trained and it's done that diffusion process millions and millions of times, we can start with pure noise, a blank canvas of static, and we can give it a text prompt, for example, a golden retriever puppy wearing sunglasses. Now step four, denoising step by step. The model removes just a little bit of noise at a time, guided by what it's learned. After each step, the image is slightly less blurry, slightly more detailed. The final step, the final image, and it's a unique creation. This continues for 50 to 1,000 different steps depending on how fine tuned the model is. After enough steps, the image becomes fully clear. You now have a photo realistic image that never existed before, created purely from noise and shaped by language. And the big question on everybody's mind is what data are diffusion models trained on? Diffusion models are trained on massive huge datasets of images, and usually they need to be paired with text descriptions, also called image text pairs. These datasets teach the model not just what things look like, but also how to interpret text prompts into visuals. Training data often includes an image, a photo of a dog wearing sunglasses, a caption or description like a golden retriever wearing sunglasses on the beach. This pairing allows the model to learn when someone says this, the image probably looks like that. The model then looks at image text pairs for millions and millions of images across its image library it uses. So where does it get these millions and millions of images? That's another huge question that brings up a lot of copyright issues, which we'll get into a little bit later. But let's talk about some of these datasets, and different models use different datasets. So there's the Leon five B, and that's 5 billion image text pairs scraped from the Internet. So anywhere on the Internet grabs whatever it can. A image that has a text description, it's go to grab it. Open Images, that's 9 million images with labels, bounding boxes, and captions. There's Coco, 330,000 labeled images with detailed captions and YF CCIOOM. It's 100,000 million flicker images with metadata. From this training data, the diffusion model learns how different objects and concepts look, what styles, is it a cartoon? Is it realistic or visually represented, how language maps to visual elements like furry, glowing, Bow house. And because some of these diffusion models use training data that is all over the Internet, there's huge ethical and legal considerations to think about. And I'm going to have a dedicated lesson to break down all the legal issues with image generation tools and ways to work around it to make sure the stuff that you're using is safe to use. So some training data may include copyrighted or artist created work, and many artists have raised concerns about their styles being mimicked without consent. As a result, ethically sourced models like Adobe Firefly are being developed using only licensed or public domain content. We will discuss in detail the legal issues with image generators in the upcoming lessons. Because each dataset is unique and each model uses a different dataset, there's different styles and properties and personalities that different image generation tools use. Example, Mid journey was super duper popular when image generation tools first came out a couple of years ago, and they tend to sample artist work all across the Internet, not as much just general Google image work. So they're kind of taking more from artists and artists work. So they're going to be highly artistic, imaginative, surreal, dreamlike. It prioritizes style over realism. And it often looks like digital paintings, concept art, and stylized magazine visuals. So if you're looking for something highly technical and scientific, Md Journey may not be the tool for you. But if you're looking for a really rough character concept to then personalize yourself, then this might be the tool for you. But it also has the most legal headache issues, and it's been the one that's been sued the most and kind of attacked by artists community the most. And we'll talk about how to avoid stealing original creators work by using these tools. There's lots of different things we can do to add our own personality to what we generate to make it unique and our own. And another example is Dali, which is going to be the hat GPT image generation tool. And I'm going to be using this the most throughout the course because I've had a lot of great results with it. And it's great at following complex text prompts exactly, and it produces clear, coherent, illustrative results. So it's good for storytelling, cartoons, editorial styles, and I found it really good for logo ideas and generation. Eonardo is another tool that's great because it have some free options for you to use. It's not as high quality as the paid dolly Open AI chat GPT image generation tool I use, but it is a great alternative that's free, and we'll talk about which options are free and which ones are paid a little bit later. But it's strong at cinematic, fantasy game style or concept art outpost, often used for product mockups, RPG assets, and UI game design elements. It's stylized yet high fidelity. Firefly by Adobe is the most approachable one that I like to use because most design students have an Adobe subscription, so it allows you access to Firefly within Photoshop, but also outside of Photoshop on their standalone AI image generation tool. I have gotten mixed results using Firefly. It could be a little tough to get it to really understand your prompt it sometimes is great and sometimes is not. And I will be using Firefly as an example, but it's kind of got mixed results. I find there are better tools, but this is the most accessible one to designers because you guys are already using Adobe tools. So hopefully they'll get better and better, and there's a reason why it struggles compared to other models. It has one of the safest datasets that it trains on. It uses images that are all copyright free. They've all been granted permission to be used on the AI tool. So you can use anything that's generated on Firefly for client work or for commercial work. Can't say the same thing for some of the other AI models. So that's what makes it unique, but it also limits its library. It's got a much smaller library to train on because it only can use licensed work where there's already been permission granted to use it. But with really good prompt writing skills, you can navigate around this issue. As you can see, there's so many image generation tools to choose from. I'm going to just focus on a handful, but my hope is to teach this class so it can be timeless. I'm gonna teach you how to do keyword writing and prompt writing. And we'll come up with all sorts of amazing creative words that really help make our prompts stand out and our work stand out. CT. C. 4. Prompt Writing: AI is not magical. It doesn't reach your mind or infer vague ideas. It interprets what you say literally and probabilistically. That's why the wording, structure, and detail of your prompt make all the difference. A strong prompt can be the difference between a generic result versus a stylized masterpiece. Cloud output versus something with emotion, texture, or story. M. Versus Wow. T. Prompt writing is creative direction. Think of prompt writing as giving direction to a highly skilled but literal assistant. The more clearly and creatively you communicate the subject, the vibe, the style, and the content, the closer the result will be to your vision. Prompt writing is not about hacking the AI. It's about communicating like a designer. Prompt engineering is the process of putting together precise and detailed prompts. Just like choosing fonts or arranging a composition, there is nuance to writing prompts. The order of the words can matter, and we know this by understanding how LLMs work to predict the next word based on the words before it. The tone of your phrasing influences mood. Modifiers like cinematic or hyperreal or dream like act as filters. We get to dig down into some very specific modifier words later on. Good prompt writers iterate, tweak and learn the visual language of the model, much like designers learn color theory or grid systems. OpenAI's President Greg Brockman shared a concise four step framework for writing effective AI prompts. This approach emphasizes clarity and structure to enhance the quality of AI generated responses. State your goal clearly. Begin by specifying exactly what you want the AI to accomplish. For example, create three distinct logo concept ideas for a sustainable coffee brand named Green Brew targeted toward eco conscious young professionals age 25 to 35. Specify the desired output format. Define how you want the information presented. For example, present each logo idea with a short descriptive name, a brief rationale, maybe two or three sentences, and suggest suitable color palettes and typography styles. Next, set constraints and guardrails. Include any limitations or specific requirements to guide the EI's response. For example, avoid overly complex or illustrative designs. Stick to minimalist, modern aesthetics appropriate for digital first branding, ensure the suggested fonts are available via Google fonts or adobe fonts. Provide contextual information, share additional background or preferences to help the AI tailor its response. For example, the brand emphasizes ethical sourcing, environmental responsibility and a sophisticated but approachable personality. The design should resonate with young professionals who value sustainability, but also style and convenience. With this, you'll get much more specific outcomes that you'll be able to gain lots of insight from. Of course, detail is a big part of prompt writing. We learned earlier how important context is to LLMs. In the last part of our four part prompt, we were provided contextual information. We can go many, many steps deeper in our explanation of brand ethos, target demographic, and desired look. We still have to do all the research for brand design, but AI gives us a springboard of ideation and lets us explore areas we might not have gone down ourselves. The interesting thing about AI is how much you have to equally partner with it to produce something worthwhile and unique. Remember, it is trained on a dataset based on human neuron networks and thought processes, so it can emulate creativity, but it doesn't know how to be creative on its own. It needs your guidance as a trained expert of design and guide its way. That is why I'm happy to report that all of our efforts, learning design theory, color, layout, typography, photography, cropping, color grading, hierarchy, design history, and styles will absolutely still be needed to produce anything of usable value. We are the art directors. We have to think of AI as a new software tool to help assist us, but we are still very much in the driver's seat. Let's move from writing for LLMs for a moment. We get to do more idea creation and brainstorming using LLMs in an upcoming brand design project. But for now, let's shift into writing effective prompts for image and video creation. I want to show you the evolution of a shoe prompt. So what are the building blocks of a strong prompt? So here's an example. A futuristic sneaker and the style of Bau house meets streetwear fashion with neon gradients, reflective surfaces, and dramatic shadows, rendered as a product showcase mockup for Instagram. Let's break it down. So what's the subject?t's a futuristic sneaker. We can be very detailed with what kind of sneaker it is. Secondly, we establish a style or influence. What art style or reference do you want? So I talked about Bau house meets streetwear fashion. Two different styles merge together. And we break it down further, we add details. What should it look and feel like include textures, colors, lighting, and mood. So for this one, we did neon gradients, reflective surfaces, and dramatic shadows. We didn't just say shadows. We said dramatic shadows. We didn't say surfaces. We really made sure we were detailed. Lastly, format medium. What format is this? A poster, logo, illustration, social media post. What we said is we wanted rendered as a product showcase mockup for Instagram. We were very specific of the kind of output and format we wanted to be in. So we start off with a simple prompt of a shoe. I didn't tell it anything but just generate an image of a shoe. There was no details, there's no context, a shoe. So what it's going to do, it's gonna come up with what it thinks is a shoe based on all the training data. It's just a white shoe. There's no characteristic to it. It's not a specific type of shoe. So let's edit our prompt a little bit. Let's add some more details. A modern sneaker with Bohuse inspired shapes. So now we have a little bit of shapes and color coming into our shoe. Okay, let's go a little more detail. A modern sneaker with Bau house inspired shapes, neon glowing borders. Put it on a black background. So now we're setting the background and the scene. We have not done that before. And now we're being very specific with what the lines need to be, which is going to be neon glowing, and it's still gonna keep that Bauhaus shape. Let's dig deeper. Let's do a modern sneaker with Buhuse inspired shapes, neon glowing borders, put it on a black background, the sneaker rests on a glossy three D rendered water with additional Bohuse inspired shapes in the background. So I'm telling more detail of the background, and I'm also telling where the sneaker rests on. It's going to rest on some water. So we're gonna push this even further. I am telling it that I want a specific photography angle. Instead of just a shoe where you see the full side of it, I want a three fourth angle, so I'm going to add that to the prompt. I want to add more detail. I want it to be raining, and I want the rain droplets to hit the water that it's resting on and the shoe and have it react. I'm telling you, not just have it rain, but I want the rain droplets to come down and be reacting. Let's layer more and more detail with this. I want some of the water to rise above the sole of the shoe and splash up against it like a wave in a storm. I want the lighting effects to be back lit with holographic properties. So let's add two more revisions. So I did the same prompt as before, but I'm adding keep everything the same, but change the shoe laces to be gold threaded. Make the gold shiny and bright. But also make it so it has slightly warped perspective. Make the gold shoe laces more reflective, add more splashes of water, lightning bolt behind it. I'm basically saying more dramatic. So lastly, I want more water splashing up out of the water. I want the water to have more reflection of the neon from the shoe. I want there to be a bolt of lightning in the background. I want this shoe to still have the boohus shapes. I want the neon glow. I want the shoe to be more idscent. Me, more, more. I'm adding more details, layering and layering, and I'm making this way more unique than it started out as. The best way to write detailed prompts and learn how to do so is by studying other examples. There are so many fantastic prompt writing examples you can find online. Right now, there is creative value for those who can write very detailed effective prompts for visual imagery and videos, so much so that people can charge for specific prompts to produce very specific imagery. It is seen now more as an art in itself, just as creative as sketching a picture or creating logo. Why? Because it takes monumentous effort to write effective prompts. The words used, how we frame the background, the setting is like speaking a new creative language, and those that know how to speak the language will thrive. That is why I put together some really cool resources for you. Before we head into that resource, I'm going to show you some real world prompt writing examples. Let's break them down. 5. Real World Prompt Examples: I found this one on Instagram. So this has this really neat fiber, embroidery, yarn look to it. And they took logos and they were able to apply this particular prompt to lots of different ones. So let's take a look at a prompt. And you'll notice the prompts that are really, really good are these big, long, chunky detailed paragraphs. So let's break it down. Create a highly detailed textured logo for brand name made of thick yarn or wool. So you're establishing the subject matter and describing it. Each section of the logo should be in a different vibrant color matching the reference image provided, and reference images are really important as well. We get to do practical projects using those. The yarn should have a knitted texture with clearly visible fibers giving a soft dynamic three D appearance. And sure the logo has a three dimensional effect with shading that makes it look like a knitted piece of fabric. So we're talking specifically about what kind of yarn, what color and how it looks. And now we're going to set the scene and the background. So the background should be neutral or light colored, allowing the vibrant yarn texture to stand out while showcasing the brand's unique identity. So when we break down that prompt, it has kind of those four distinct layers that we talked about earlier, where you're establishing the subject, you're giving it context, you're giving it a background information, and you're giving it in the format that you want to have it in. This next one I really could have used in my graphic design intermediate master class where I taught you guys how to create a fast food poster, and I had to go to pexels.com to find a free photo. But it was very limited. I ended up finding something that works. But what if I can create something that perfectly match my creative vision for the poster? And this would be really neat to do for any kind of fast food poster or any food related items or any poster you want to generate. Now you can create whatever you want in terms of using that photography. So you can also do tacos, different kind of food objects. Prompt isn't as long, but I think it's still effective. Says, render a dramatic hyperrealistic image of, you know, whatever food suspended in midair with crumbs, splashes, particles frozen in motion. Use bold rim lighting, macro focus, and a bright, whatever color background to add energy and contrast. So you notice some of these creative keywords that when they are dropped, it really gives the AI sense of style to look for in its references. So in this case, rim lighting, macro focus. And suspended in midair and hyper realistic. You might not know a lot of those words, but we get to go through some really nuanced words to help us be able to come up with those really strange words that maybe we don't know what rim lighting is. But we get to explore all that here soon. And I love seeing these hyperrealistic textures being used, especially this inflatable blown up kind of object here that you're able to do. So let's take a look at this prompt. We can see it in action. It's very consistent. Once you develop that prompt, you can just kind of change out the subject matter, and it's gonna keep that same style pretty consistent. So how are we going to create this? This is how they did it. So let's take a look at the final prompt. Wow. I'm not going to read all of this, but I wanted to show this example of how elaborately written this is and how it probably took a good solid couple hours just to write the prompt, see the results. It's not what you wanted. You got to tweak it, change the prompt, just like we did with the shoe example. We had to go back, add stuff, add stuff, add stuff, add the details. This is hours, and this is why writing prompts is a creative art unto itself. So let's highlight maybe some of the really nuanced specific keywords here that they're using. So they use inflatable transparent object. Gently floating in the water, so it's not in rough water, so it's kind of setting the background. So it has smooth bulging surfaces, thick, visible, heat sealed seams. So let me tell you how specific that is. So you almost have to research how plastic objects are made and terms about plastic. So sometimes you have to go to hachPT, search about inflatable plastic, learn about the process to even know how to write a prompt on how to emulate it. So that's how detailed this stuff gets. So you have turbulence, air bubbles, faint ripples, soft natural caustics. So we'll learn about caustics in a little bit, but that's all about lighting. I didn't know about this until I started to really dig deep on keywords for writing prompts. Here's another fine example of prompt writing. This is in a vacuum packed sealed bag. So let's kind of see how to do this effect. So create a high resolution hyperalistic image, and you'll see that over and over, these same keywords, hyperrealistic, high resolution because those are these little keywords that all of a sudden click in the AI model's brain that goes, Okay, this is the type of photos I need to look for. But I wanted to go down into this little area, include condensation or small creases around the pressure points for added realism. How beautifully written is that? So at the end, you can see this visual details with a colon. So it's going to list lots of visual details more than what it's already done. So crushed, transparent or metallic vacuum plastic, object silhouette, visible and extreme detail, harsh lighting to emphasize texture and form, typographic overlays, skew codes, and branding moody, product display style, background mood. So it's setting the background mood, experimental, edgy, collectible post consumer bright natural lighting, and enhances the vivid colors and gives a clean, cinematic, realistic look. Beautifully written and the prompt, of course, looks fantastic. I had to try this prompt out. This is what I got. So this keyboard example was really neat because I thought that they really described what they wanted with those extra keywords. So in this case, they said they want a tight two by two grid. They just didn't say, show me a keyboard. They said, I want a two by two grid. So two keys on the top, two keys on the bottom. And also another thing is they talked about isometric angle. So that is the camera view and focus. It's got this isometric angle. So if you guys have studied, I've taught you guys isometric design before. So that's kind of really popular in terms of the view. So uploading reference images is really neat. So they uploaded a reference image of a photo they took of a Coca Cola can, and they added the prompt a high resolution image of this object floating inside a couple of white clouds casting shadows in a bright blue sky. The chrome slightly scratched, dented, but highly reflective, bright energetic lighting with a surreal, dreaml atmosphere. And you can see how you can make a lot of really neat mockups with your own products or brand design work that you're working on. For this last example, it really inspired me to do some keyword searches for similar lighting, textures, and terms. So this has this iridescent look to it, a really, really cool effect where you have almost this rainbow prism, reflection and glass. And if you don't know if your prompt, your road is good enough, you need a little extra boost added to your prompt, simply ask Chat chPT. It's great at refining your prompts a bit more as it knows the types of prompts it needs to generate the type of content you're looking for. Obscure descriptions unlock uniqueness and style layering. Most of us designers get stuck using the same ten or 15 visual keywords, perhaps bold, thick, bright, geometric, round to name a few. But what if there was an entirely new world of thousands of different descriptive creative words that we've never explored before? It all started when I saw that iridescent prompt that I showed you earlier, and I thought the effect looked really cool. I've seen it before. I just didn't know how to put it in words. I'm going to be honest. I did not know the difference between idescent, luminance, bioluminescence, and all these other different ones before researching this class. What's incredible is I can type into chat GPT or a similar AILLM and ask it for similar words for idescent. It came up with a wider variety of similar words. I was able to ask it to create visual examples of that particular lighting on the same object, so I can get a sense of the nuanced differences between the different lighting effects. Wow. I would have never dived so deep into such specific words before, and now I feel like a better designer. I can now deeply describe various different lighting situations when I write my prompts. And also ask Cha GPT a very specific nuanced art styles, textures and moods, so I can expand my designer's vocabulary. I was able to take this list and intensely research so many new varieties of textures and art styles. I feel like a brand new designer that has the whole world at our fingertips. Of everything I've used for AI, this is the one that kept me up at night. But in a good way, I want you to personally go down the rabbit hole of exploring nuance design terms you may have never heard of before. That way, when writing prompts, you can be insanely specific. We'll go over lots of examples of nuance design terms in the next lesson. So get ready. O 6. Nuanced Design Terms: A biometric sculpture and a guessoed texture under gloaming light with Wabi sabi sensibilities. How on earth do we get to learn what all of that is? I don't even know half of those terms until now until we really start to explore some of those nuanced keywords that can make our prompts really professional. So I have this downloadable resource. That's what I'm going to be looking at with you together in this class. So when writing prompts or developing creative briefs, these words reflect lesser known our nuanced styles, textures, and aesthetics that can elevate your design language. So let's work on expanding our design language. So these are some and, of course, chat GPT and I really work together to fine tune some of these very different styles. So I want to talk about the first one. These are some different aesthetic styles and movements. So a couple of ones I want to show you the biomorphic, which is the example in the beginning of the lesson, kind of this organic, blob like fluid forms around in nature. And I was able to deep dive, I Googled it, researched it, and now I know biomorphic design, which is, Hey, I can develop some three D biomorphic elements to put on a brand design project and tweak them because I really like how that looks now, especially in a three D model. And there's y2k core. So I lived through the early 2000s, so I knew very much about y2k core. It's a retro futurism form of the early 2000s, Chrome gradients, chunky tech. So think web two point oh with that glossy kind of look vorticist. It's angular dynamic abstraction, industrial modernity. So this was kind of a popular painting style, and I was going to look at lots of stuff on Wikipedia about this style. Really cool. Didn't know that existed. Dynamism dynamism took me a while to figure out how to say that word. It's high energy compositions. So let's move on to different textures and material descriptors. So here's that iridescent. It's a shimmering rainbow surface, a color shift with light. I thought that was so much better than just saying neon. I felt like the only way I could describe a glowing light was neon. But now I have dscent. So you have oxidized, which is when you have iron that rusts. So it's a rusted chemical patina with weathered metals. So this can give you that worn metallic look that maybe you were looking for. There's also fleck, which is scattered particles or sparkly texture. And then there's de collage. So D collage is the torn away layers revealing a visual history. And I've seen this in designs a lot where you have that torn look and you see the layers, and really loved how that looked. I just didn't know it had a name D collage. So unless you took a lot of art history classes in college, you might not know some of these, but you can have Chat GPT help you now. Then there's the scary, grotesque, intentionally awkward, distorted and uneasy. This one was kind of creepy, but hey, we all have different things that we're creating for our design pieces. So one of my favorite is lighting and mood terms. Caustics is actually a very popular term in video and three D model rendering, talking about how light interacts with water. And sometimes glass as well. So it's refracted light patterns, often underwater or glass. I always love this effect. I would hand paint some of this when I used to be able to do digital painting and I would paint water. I'd paint that kind of reflected wobbly kind of lattice, and that is caustics. That's the study of light and refraction in glass and water. So need to discover that and to be able to put a name to that. Let's move on to cultural and niche style terms. Retro futurism, which is a vintage sci fi visuals, imagining the future from the past. There's Wabi Sabi, which is a Japanese concept of imperfection and transients. That's why you'll sometimes see the gold cracked repaired and plates because they want to honor the imperfection, and it's a very cultural thing to honor. And I've actually seen this Wabi Sabi in a lot of prompts lately, so it must be a pretty popular style. So there's also aesthetic and visual compositions. So orpism is a vibrant abstraction using color to express musical rhythms. So rainism is intersecting rays of light, semi abstract futurism, and even fractalism, which have you've heard of mathematical fractals, which just go on for infinity, are recursive geometries, self similar shapes and complexity. So then one of my favorites is surfaces, patterns and media techniques. Here's one vertigris which is oxidized copper, and it kind of gives us really cool green patina. And then there's color behavior and effects, prismatic color. It's kind of like when you have the prism, which is basically the whole rainbow of light kind of being refracted in kind of a tight area. So you have a rainbow refraction with sharp transition. So they don't have these loose transitions like loose gradients. They're very tight, and you see very quick transition in color. This frenzl lighting, and I hope I'm pronouncing all this correct, I can always a chat GPT. But frenzl lighting is high energy reflective gradient at edges. So it's got this cool gradient just around the edges with everything else being kind of dark and unreflective. And then we have soap culture and global influence. So you have desert modernism, which is mid century architecture adapted for arid climates. So if you ever doing a prompt for any type of building or a building in a background scene, you got to them tell chat GPT, or whatever image generator you're using. Make sure that you put in what type of building everything is in the background. Be very specific of what style of building it is. And then another super, super common keyword I'm seeing everywhere is Neo Tokyo. And Neo Tokyo is this gritty, colorful, anime influenced urban sprawl. And I'm seeing this a lot when you have futuristic robots. They tend people just want to put that in a Neo Tokyo setting. So you know me. I like to go further and further and further into the rabbit hole because I have such an intellectual curiosity for all things creative. So I asked ChahPT What are even more nuanced, rarely known terms in design and art. So if you find a style that you're like, What on Earth is that, you can upload a reference photo to an LLM and ask it to describe that style with prompt words, and it's good to help you figure it out. So a couple of my favorite rare terms was flocked velvet. I just feel like I can reach out and touch this. Leucite plastic, which is a retro plastic, kind of this thick, chunky plastic that was really popular before they started getting the really thin plastics we see today. But I can see this a lot in, like, retro products. Bubble wrap texture. I love how this looks on stuff. Subsurface glow, which you can imagine, like a block of magma and just a little bit of the magma from inside is coming out, but it's not super bright. It's very subsurface. And ferro fluid shimmer. So have you ever seen oil spilling on a road, and you see this rainbow kind of reflection off of it? That is exactly what this is. It's this reflection that oil gives off where it kind of reflects a little bit of the color spectrum at you, but still has that dark, liquidy look. Let's go even crazy deep. So these are probably ones you have never heard of. And if you've heard of them, congratulations, 'cause I haven't heard of them. So these are the most obscure prompt words I could find on the Internet, and the amount I found was endless. This is just a very small selection that I personally liked and I thought you would find useful, but there's thousands that I didn't use. So there's moon glow refraction, a soft, silvery light bent through mist. Very, very specific. There's kaleidoscopic bloom. So if you ever looked through a kaleidoscope, it's chaotic fractured light dispersion, photonic bleeding, which is overlapping light sources with a noisy overlap. There's magma polished stone, which is sleek and scorched, cooled lava meets obsidian. Vitreos bark. I think I'm saying that. I think it's from vitae meaning life. Vitrios bark is tree bark with semi transparent glassy sheens. And there's these ultra trendy conceptual themes we can talk about as well, Archetypal glitch core, which is broken, symbolic language. Forgotten utopia fragments, broken pieces of failed perfect societies. I mean, how nuanced can you get? And then solar punk ruins, which is echo utopia that has already decayed. So if you're trying to paint a dystopian mood board, then these are definitely some words you might want to use. As a secondary student challenge, I want you to find the most nuanced art styles. Pick four different nuanced art styles, textures or descriptions, and I want you to explore that intensely. So if it's iudescence, look up iridescence, find out what that is, learn about it. And I want you to do that with four different ones. And if you want to do two every day for the whole time you're learning AI or doing this class, that'd be great because you can really start to expand your vocabulary and design knowledge that way. So now we understand the basic framework of what it takes to make a solid prompt. And also figure out some of those nuanced keywords. Ix, there's just one more thing left to discuss the legal issues with using AI. This has to be talked about before moving forward. 7. Copyright & Legal Issues : So where does AI source its photos to create such masterpieces? It's hard not to talk about the elephant in the room. As we discussed before, Mid journey Dali and other AI photogeneration tools took a huge swath of photos from the entire Internet to train its AI bots to generate images. That means that copyrighted photos, illustrations, and graphics were compiled together to teach the bot what the user might want to see. There is an interesting article that claims that one of the founders of MD Journey knew this was the case and admitted to not knowing what to do about giving proper copyright ownership to the artists of the images this AIBt uses. When creating AI art, you can also add reference images to help the bot further detail what you're looking for. And there's no way to prevent users from uploading copyrighted work from Google Search into the prompts. That means if you're using images that do not have a creative commons zero license or a public domain license, you could be opening up yourself to being sued for driving artwork from copyrighted images. So does that mean AI tools have infringed on creators rights? This was going to come to a head at some point. Several artists have banded together to sue Mid Journey in other art portfolio websites like Deviant Art for allowing copyrighted, derived AI work to be posted without giving proper credits to the authors. And it's going to be a very tricky court case. On one hand, AI tools have been trained by absorbing data from most of the Internet, which is a gigantic source of data. It could be hard to prove individual copyright infringement from images derived from such a large dataset. On the other hand, there have been cases where individual artists can type the name of an AI prompt and clearly see how their artwork was used to formulate the results. Albeit, it's not ever an exact copy, but you can see the inspiration. Who owns the work created by AI image generators? If I put in a prompt into an AI text or image generator, do I own the prompt to create the image or the image itself? It's a complex legal issue, but it's always worth reading more about this. A human element has to be present for any copyright claim to take place. That means AI Tech cannot claim ownership of images. AI artwork does not really have an owner based on current copyright laws, but according to the terms of use of some of the programs, it does assign the ownership of an image to the creator or prompt writer. But can you hold that copyright claim in the court of law would be the next question, as nothing can stop third party companies from taking you to court for using their brand image in your AI generated photo. We are truly living in a new digital Wild West. So what do you do if you want to take the safe and high road and protect a real artist's work and make sure they get the proper credits? Well, first of all, I would avoid putting in a specific artist names into AI prompts. It's okay to use historic names like Leonardo Da Vinci. He's been dead for many years, but I wouldn't put any new artists that are still alive and still have a legacy to build. Another thing you can do is to make sure you use AI image generator tools that are from official companies that make sure the library of photos they use to train their bots and to generate images are given permission by the people who own them. Outside of using AI tools like Adobe Firefly, here are some personal best practices to mitigate against these complexities and ensure your work is unique. First of all, each AI tool has different licensing terms, so it's good to review each one. Have an LLM, break it down for you so you can digest and compare the different terms. Document your creative input and iterative steps clearly. Let's say you're doing a character design. Perhaps you keep a copy of your original sketch that you uploaded to an AI model. Keep track of the different prompts you use to edit and change your character. Use caution when prompting AI with copyrighted characters, famous brands or celebrity likenesses. This is where you can get into the most trouble and have the highest chance of getting sued. Avoid using company names when typing in your prompts. Avoid using Nike logo to generate ideas for a logo. Avoid saying Pixar or Disney animation style when creating images. A prominent example of this is when someone started to copy the famous Ghibli style by famous animator Hao Miyazaki. His style takes hundreds of hours just for a few seconds of animation, and people were inputting his name and style into prompts to emulate this animation. In an interview, Miyazaki called AI an insult to life itself, and he believes that animation should be rooted in human emotions and experiences and not algorithms. So when writing your prompts, think about creating your own mix of styles that will be unique to you. You can be inspired by other creatives work. We do this all the time as creatives. We browse Instagram our Behance for inspiration. We then go on to create something, and we find ourselves emulating some of those styles subconsciously. The same issue exists for using AI. Finding originality can be difficult, but that has always been the case for us designers. This is why we study hundreds of styles so we can mix, match, and create our own unique flavor and factor. Ways you can establish your own style is to upload a basic sketch of your idea, logo, or character. Writing very elaborate prompts that only could be written by you. Maintain a specific style in what you generate, which allows you to take ownership of that style. Taking AI generated ideas and modifying them heavily outside of the AI programs and design programs like Photoshop and Illustrator. I recommend a back and forth handing off of creativity in your workflow. That means you might upload a rough sketch into AI. It helps you refine your image. You bring that back into Adobe Illustrator or another vector program, and you vectorize it. Then you can tweak it further. You bring it back into AI to add additional details or refine the ideas. You can even ask the AI for advice on direction for your logo afterwards. That brings up the issue of getting sued. How likely are we to get sued for using AI generated images in our marketing campaigns, for instance? It's possible but not very likely. The person or company suing needs to prove without a doubt that the images you use copy their style exactly. Since AI generators are trained on millions and millions of text image pairs, that means it's impossible for one photographer or designer to claim ownership if your prompt is unique enough. But one could write a prompt that describes a famous photographer's style to a T and generate it to be so close in that style that it infringes on that person's style. This is really tricky. It will always remain a gray area as it still does with logo design. One thing you can do is regularly check AI generated assets using reverse image search to detect potential similarity or infringement issues. You can modify AI generated output significantly rather than using them directly. So if you ask for a logo prompt idea, modify that prompt just a tiny bit to make it your own. Post processing, you can edit AI generated images extensively using tools like Photoshop or Illustrator. In the end, the big takeaway is the more human guided interaction there is between the AI generated content, the better you can protect yourself, and short clear ownership depends significantly on how much original creative human input you add to the AI generated imagery. Always enhance and adapt images creatively to clearly establish your copyright. Ensure your final designs contain meaningful creative human modifications and are free from infringement concerns. 8. Student Project: So I have your first student project, and that's to reverse engineer a photo using prompts. So I want you to recreate the reference photos as accurately as possible and you can download these as part of the resources using only AI image generation tools and written prompts. No manual image editing allowed. This will allow you to practice using prompts to create very specific objects and details. So here's the first one, which is three D shapes. I want you to recreate this as close as possible. It's not going to be exact, but you're going to be able to continually edit the prompts so you can slowly get the results that you need. And the second one is an American breakfast. So you could start off with two eggs and describe everything in the background, describe that there's a window, describe that it's photorealistic. Start using some of those keywords that we've learned about to be able to describe this picture and to emulate it. So choose any AI generator of your choice. It could be Adobe Firefly, Dolly, Leonardo, hat GPT. Use only text prompts and attempt to replicate the image, so you can't do any reference images. And I want you to get close to the composition, the lighting, the subject, the color palette, the texture, and the style. And you may iterate as many times as you needed and keep track of your best prompt versions as you were fined. 9. BONUS! Nano Banana Pro - Can You Guess Real Or AI?: Google Nano Banana Pro just came out, and the results are insane. The ways in which AI leave evidence that it's AI is slowly becoming harder to spot. It makes cheating, changing the past, scamming, and fooling people much easier. Of course, it has its benefits for those who know how to use it correctly. For Photoshop 2026, Adobe announced it's partnering with Google to add Nano Banana as a third party option in its generative fill tool. Now a new option is available in its Nano Banana Pro. This is a paid option if you were to use it in the Google Gemini, but Adobe made it available to use Adobe Photoshop right now. I've never seen such amazing results with this option way better than even the already pretty good first gen Nano Banana. Since it is a premium option outside of Adobe, Adobe does penalize you a bit for using this Dano Banana Pro. It costs ten credits per generation for one to 2000 resolution and 16 credits for 4,000 resolution. Creative Cloud Pro gives you 4,000 generations per month. Adobe Creative Pro does cost $70 a month in my area, so you're definitely paying for it. So you might as well use some of those credits and experiment with this tool. So I want to play a little game with you. I generated most of these images using Google Nano Banana Pro with the exception of one or two. I want to see if you can tell me which one is generated with AI and which one is a real photograph. I did this with my husband, and he failed miserably. Let's see how you do. Please write in the comments, how many you got right. I will let you know how to tell the difference between real and AI after each. Now, we're going to start off with one of the easier ones. This is one of the only ones my husband got right. Sorry, honey. So which one do you think is real and which one do you think is AI? Just give you a few seconds to think about it. And it's all about zooming in. So if you're not able to zoom in, it's very difficult to tell if it's AI or not. From first glance, I would think maybe the one on the right. It looks almost too good, but the lighting, that's really nice. And I'm not an astronaut, so I don't know if her equipment is correct or not. It seems kind of complicated in the front. But then again, the one on the left seems very convincing. But let's zoom in and find out which one's real and which one's not. So if we zoom in to the one that looks like it's taken from the 80s, it's very believable because there were women in the early 80s that were starting to train in the astronaut program. But take a look at this badge. Typography in Texas where AI still struggles, even with Nano Banana Pro. It's the only thing I can ever catch it with 100% accuracy is when it comes to typography. It's not absolutely crisp and clear. You can tell with a NASA badge. Also, up here, you could tell there's little holes, almost like it's worn, but I can tell it just struggled with the typography, and that's not the official logo. Also, if you look at this guy's face, they must be twins because it's the exact same guy. So if you have the same exact guy, what are the odds of having twins in the space program? Probably very minimal. But it's very impressive. Look at the ceiling. Look at the wood paneling. The details are quite good. And at first glance, I would think this was a real documented photo. The one on the right is a real photo, and the one on the left is AI. Were you surprised? Let's go to the next one. Let's get a little bit harder with this one. So here you have two standard movie sets, one with Tom Cruise on the left, and the other one with Pedro Pascal and some other famous actors. Which movie set is real and which one is fake? And this might be a trick question. So I want you to take a look, and they look really convincing. So let's take a look. The only way to really tell is to zoom in and look at the textures. So this one's super convincing. Look at this camera work, look at the green screen, look at the sky. You can't really tell. Look at this building. It's not warped. It's not distorted. This looks surprisingly good. Take a look at this flooring. There's not any mistakes in the flooring. If you look at the shoes, you can't really see too many errors until a ha. Look at this edge here. That's a total AI generated image. Also, the more you zoom in, you can see this repeated texture, almost like a computational texture right here. When you zoom in across all of the tile. That is not natural. That is not a natural texture. So all of a sudden, when you zoom in, you start to see all the little mistakes. Maybe he has a missing finger, but you could just say, Well, it's bending one way. So when you zoom out, you can't really tell, but when you zoom in, you can. There's also actually, this was a photo posted on Red it, like a real or AI red it form. And there was a lot of professional camera people that pointed out tons of issues with this camera. So when you are in the business of movie making, you could see tons of mistakes. Also, someone pointed out that this green screen isn't exactly the same angle appearing here that it actually is showing. And the biggest giveaway for me, or what most people would be able to spot is up here in the scaffolding. There is some really strange wiring and bending of the wires. So that was a dead giveaway. But when you zoom out, this is so convincing. But when you zoom in, you can always tell. What about the one on the left? This one's pretty convincing. I mean, he does look about his age. Maybe this is a 7-year-old photograph or a 10-year-old photograph of him filming something for Mission Impossible. Well, wow. This looks pretty good. The little details left. See this man holding the photo. We have the wires, a lot of natural human like details left. You can even see them snacking on set with biscuits and coffee and even have this little logo on his jacket and even a pin. Is all seems super convincing until you realize this one's also AI. Once again, the camera that camera doesn't exist in real life. You could do a reverse image search, try to find this camera, and you won't be able to find it because it does not exist at all. But this one was probably one of the harder ones. There's not this Omega obvious way that it's AI, but as you can tell, Google nana, banana really rocked this in almost a very scary way. So yeah, they're both AI. But the one on the left is actually way more convincing when you zoom in. So really scary. Alright, all you creative people. Let's do one that applies to you. One of these is a real sketch, and one of the is AI. Which one is which? I'll give you a moment to think about it. Look at all the details, and I'll zoom in on some of these so they can see it. Well, they both look like sketches. Let's take a look at this one on the left. It could be AI, but it's hard to tell. This looks pretty genuine. If it's AI, I'd be really scared. There's lots of leading lines. There's lots of human kind of experimentation with trying to figure out the shapes of the typography. I don't know. It's pretty close. How about the other one? That looks really convincing with that worn down chewed up pencil. But is it too chewed up? I mean, who actually has a pencil that's that bad? What about that looks like a really, really, really, really oddly shaped eraser. But the sketches themselves are a little too crisp. I see these leading lines, which makes me kind of think, Well, maybe this is real, because those leading lines, how can an AI do that? You know, that's only something a sketch person does to try to figure out the symmetry. Either this is a really talented sketcher or it's too symmetrical. Even the little dust that's left by the little pencil dust or eraser dust is really convincing. So which one is which? Okay, so the one on the left is actually a student of mine, Amber Axelton, she did this as part of a branding project. So the one on the left is real, and the one on the right is absolutely AI. It's got kind of a super dark tone to the sketches. So usually pencil doesn't have this dark color, and only someone who's sketched a lot in their life will kind of be able to identify that. And that's going to be the issue with AI is only industry experts in what you're looking at can be able to go, You know what? I think that's fake. It just looks a little too polished to me. Who doesn't like a good game of chess? One of these is real, and one of this is generated by Google Nano Banana Pro. Which one? I'll give you a few seconds to figure it out. Yeah. Okay, so at first glance, this one on the right seems very AI generated. It's super polished and has this hyper reflection, almost like it was generated with a really good three D program like blender. And when I zoom in, I can see some details of the horse that looks very weird, and it does remind me of AI. And let's take a look at the other one. The other one seems kind of natural. I'm seeing kind of some wood pieces that are kind of chipped off. It just has a lot of natural texture. And if you see how the light is hitting it and reflecting, it seems very, very, very natural. And the pieces seem like they have organic natural texture. There's even a background that looks convincing with some coasters, some stacked books, and a coffee cup. So which one is real and which one is ahi? This one fooled everybody I tried. So the one on the left is actually AI generated by Google Nano Banana Pro, and the one on the right is a photograph. It's been brightened a little bit in Photoshop, but it is mostly an intact original image. Does that surprise you? Does that shock you in any way? Did you get fooled? So here we have two seemingly normal pictures of fruit, but one of these is fake AI, and one of this is a real photograph. Now, which one do you think is AI and which one do you think is a real photograph? I promise, they're not both AI. They both look like AI. But let's take a look at this one on the right. It seems super glossy, almost a little too glossy. It almost feels like it's glossy for no reason. It just has that extra shine that feels a little bit artificial. The table itself and the texture looks pretty convincing. The bowl, nothing else is really misshapened. This could be real fruit. AI has a hard time with stems and finding out where those things go on fruit. It's kind of convincing, but that glossiness is throwing me off. Okay, what about this one on the left? I mean, I think the way it was maybe taken in the 80s or the late 90s or something. It's got a very convincing refrigerator. It's got some medicine on the countertop. A lot of things you would just see in a random kitchen. Even the magnets are intact. There's a piece of paper on there. It all looks convincing. The shadow's correct cause the light is coming from the top and shining down. And it even has this August 14, e 96. It would really match up with a in 96 kind of kitchen. So let me zoom in and see if I can see if this is AI or real. Oh, I think I found it. Look, see this little sticker? If it wasn't for that sticker, I would have a hard time figuring out if this was AI or not. But once again, stickers, logo, badges, and typography is really hard for AI to generate. So that does not look natural. But everything else does, and that is really tricky. So the one on the right is a real photograph, which I found on Wikipedia, and the one on the left was generated with Google Nano Banana. So which one of these photos is real and which one is AI? They both seem like they were taken in the 50s or 60s. They're on a car. The one on the left seems super oversaturated for a photo, but it could have been color corrected. The one on the right seems pretty convincing. But which one is AI and which one is real? Okay, the one on the right is generated with Google Nano Banana. So it is AI, and the one on the left is the real photo. Are you surprised? Did anything about the features of the faces trick you into thinking it was AI? So, this one is a little bit different. This is a real birthday picture of a cake I took, and it's of my son and my niece's birthday. Did they turn nine and seven, or did they turn eight and six? That would be the big question. One of these is modified using Google na banana, where just one element was changed, which would be the birthday candles. Now, which one is more convincing? I think the tricky part about this one let's zoom into this one on the left. So you have this kind of weird looking thing with the nine happening, almost like it's broken, and it doesn't seem supernatural. So my first instinct would be say, Okay, this is the AI generated image. And let's go over to the other one. So this starts to look almost too polished. If you look at it, there's almost no mistakes. Not even a little bit of texture from the wax of the candle. That makes me think that this is definitely the AI generated photo. And if that's the one you guessed, they were turning nine and seven, you are correct. That's the real photo. And the fake one and the modified one is the eight and six. It is scary how we can change photos from the past and give people an entirely different narrative of what happened. This could be really scary. We can say that we dated people we never dated, or we have a receipt for a product we never bought. This is the scary part about AI. One more bonus round just for fun. Is this a real photo, or is it AI generated by Nano Banana? At first glance, very convincing reflection. Look at how everything is reflected. That seems very authentic and real. Look at how it's raining outside, and you can see the rain on the streets. The cars, they look pretty normal, although is that the same car back to back? I don't know. I'm starting to doubt myself whether it's real or not. Her hair seems convincing. The lighting, everything seems really natural. Look at the detail of the kitting, too. I can't find a mistake in the knitting. But as always, let's find any type of typography or font or text that we can. What is berating sweet grass? Don't know what that is, but that typography does not look natural. Everything else about it is too small for me to read, but that is the giveaway. That's honestly the only giveaway in this entire photo. You can zoom in and look around yourself, but it is hard to find another reason. So if she wasn't holding a book, if she was holding something without text, this would be very hard to find out if it's real or AI. But yeah, this is AI. I hope you enjoyed this little deep dive into trying to figure out what is real and what is fake. So it's getting harder and harder and Google Nano Banana, the stuff I generated from there with very simple one sentence prompts was quite incredible and quite convincing. And I could see in another year or two, we're going to erase all those little AI artifacts, and it's going to be very difficult without having a super trained eye. So continue to train that eye, zoom in at 800%, and look at those fine details, textures, patterns, consistencies. Definitely typography. Please leave a comment below if you enjoyed this and how many you got right? Or if you got any wrong, are you surprised about any of them? See you in the next video.