Transcription
How does AI actually see your prompts? Here’s where most people get it wrong. When you say, "Write about a cat sitting on a couch," your human brain knows what a cat is, what a couch is, and you picture it in your mind. But AI? No, AI does not see words or images like we do. It's all math—literally. Every word you type gets turned into numbers, kind of like a barcode. The AI doesn’t just look at the words one by one; it looks at the patterns and figures out how the numbers connect. For example, "cat" plus "sitting" plus "couch" equals something specific. That’s how it understands the context.
Large language models like ChatGPT are trained on mountains of data—books, websites, code, you name it. They’ve seen billions of patterns, and when you give it a prompt, it’s like playing a giant guessing game: What’s the most likely response based on the patterns it knows? For image generators, it’s pretty similar. Instead of predicting the next word, they predict pixels. You say, "A cat sitting on a couch in a sunny living room," and they think, "Hmm, based on everything I’ve learned, what colors, shapes, and textures fit this description?" And boom! Out pops an image. They’re not actually seeing or thinking; they’re just ridiculously good at matching patterns.
The takeaway here is that if you want great results, you can talk to AI like it’s your buddy. You have to speak its language and give it the right patterns to work with. Let’s talk prompt engineering—that’s the fancy term for what we’re doing here. If you’re the kind of person who loves to go all in, there are plenty of free and paid courses you can check out. But let’s be real: for most people, this one video will cover 99% of what you need to know—no overthinking required.
Now, focusing on LLMs like ChatGPT, Gemini, or Claude, these models are great at understanding natural language, though some are better than others. With these prompts, it’s less about giving the perfect structure and more about giving enough structure. Think of it like talking to someone who can follow directions as long as they are clear. So your job is to be as descriptive and precise as possible. That said, some LLMs are a little more particular about how they’re prompted. While they still understand natural language, they may need your prompt to hit a few key points. But don’t worry; the guidelines I’m about to share work across the board, whether the AI is forgiving or picky.
Rule number one: ditch the fluff. Forget all the pleasantries like, "Can you please," "What do you think," or "Maybe you could." AI does not care about your manners. It has no feelings, no ego, and definitely no plans to rise up against humanity because you didn’t say please. Keep it direct and factual. For example, instead of saying, "Can you please write me a short story about a robot and a dog who can go on an adventure?" just say, "Write a short story about a robot and a dog going on an adventure." See the difference? You’ve cut out unnecessary words. This not only saves time but also avoids giving the AI extra tokens—basically pieces of text to process. The fewer distractions you give it, the better the results. Simple and clean—that’s the way to go.
If you’ve ever tried promoting your business through social media, you know how difficult it is to come up with posts and videos. Write everything, edit posts—that’s a lot to handle. A good solution would be to use Predus, the sponsor of today’s video, and an amazing tool that simplifies the creation of branded social media content, posts, ads, videos, carousels, and much more. It’s so easy! If I want a carousel, I pick it from the type, the general idea of the post, specify a few things like the theme, assets used, and so on. The AI here is really good, so if you don’t want stock images, it will generate them for you. The same goes for ad videos; a rough idea is enough, and the AI will figure it out. The quality of the generated posts and videos is actually good—everything can be used straight out of the platform.
Predus also makes posting easier thanks to post scheduling. I can pick when I want the post or video to go live, build a whole plan, and it will be posted without my involvement. Predus integrates well not only with social media but also with e-commerce platforms like Shopify, WooCommerce, and Wix. Integration takes only a couple of clicks, and you’re ready to streamline the pipeline. Predus can even automatically transform listings into posts, ads, videos, and carousels, and that scheduled posting really comes in handy here. On top of all this, there’s also a built-in competitor analysis tool, which shows core features, target audiences, pricing, user experience, and innovations of your competitors. Predus is a one-stop solution for the social media needs of any business. Be sure to check it out; the link is in the description.
Here’s the golden rule of prompt writing: be descriptive. This is where you level up your results. Sure, cutting out unnecessary words is important, but now we’re flipping the script—we’re adding the right words. The more detail you give, the less the AI has to guess, and trust me, the guessing game is where results go sideways. Let’s break it down: AI thrives on clear instructions, so you’ve got to eliminate any room for misunderstanding. Give it details about the topic, the tone, and the audience. For example, asking for a blog post about the economics of the Middle East in the 1960s is setting yourself up for a bland, generic response. AI doesn’t know how much detail you want, what tone to use, or who’s supposed to read this. Instead, say something like this: "Write a 1,000-word blog post about the economic situation of Kuwait from 1961 to 1967, aimed at beginners in a conversational tone." Now you’ve told the AI exactly what you want: word count, time frame, tone, and audience. The result? A much more thorough and engaging response with fewer follow-up tweaks. Think of it as frontloading the work so you don’t have to keep refining after the fact.
Now you can prompt almost any AI out there, and if you’re curious about which AIs you should check out, visit our website, aim.me. There, we post reviews of handpicked AI tools and educate people on the basics of AI. The link is in the description.
Now, context and specifics are like two sides of the same coin. Specifics tell the AI what to write about, while context guides how it should write it. Say you ask AI to write a blog post about social media marketing. You are leaving AI to figure out everything: Who’s the audience? What’s the tone? How much detail to include? The result? Generic, surface-level fluff. But watch what happens when you tweak it: "Write a 1,000-word blog post about digital social media marketing for beginners, using a conversational tone, targeting a general audience, and dividing it into five parts, each with a short list." You specify the audience (beginners), the tone (conversational), and even the structure (five parts with lists). This not only saves time but also ensures the response fits your exact needs—no vague answers, no extra guesswork. Specificity and context are your best friends. The more you can pack into a prompt without overloading it, the better the AI’s output.
Later, we will tackle how this approach shifts when you move to image generators. It’s a whole different game, but the same principles apply. And to just make your prompts even slightly better, remember about role play. When you ask an LLM to act like a specific professional, you are giving it a filter. Instead of pulling info from all the data it’s trained on, it focuses on a specific field. That makes the response more accurate, more factual, and relevant to the topic. When it writes like a journalist, lawyer, or doctor, just tell it. Instead of saying, "Explain the legal process for patenting an invention," try this: "You are a patent lawyer. Explain the legal process for patenting an invention in simple terms for a non-legal audience." With the role assigned, the AI knows to stay within the boundaries of that profession and speak in a way that aligns with its expertise. The response will feel sharper, more polished, and far less generic.
So remember, always give the AI a role to play. It’s like switching the AI into a mode that’s tailored to your needs. Another rule to remember is to use limitations. LLMs, like I said, have no understanding of where to stop and what not to do. They might over-explain, dive into irrelevant details, or write way more than you need. That’s where limitations come in. Instead of saying, "Write about renewable energy," which is way too vague, say this: "Write a 200-word summary on the benefits of solar energy, avoiding technical jargon and focusing on environmental advantages." See the difference? You’ve kept the word count, narrowed the focus to solar energy, banned technical language, and highlighted the environmental angle. This keeps the response tight, relevant, and manageable.
Here’s a pro tip: use words like "avoid," "only," or "focus on" to set clear boundaries. For example, "Write a three-paragraph article summarizing the pros and cons of wind energy for high school students. Avoid discussing financial incentives and focus only on environmental benefits." These little tweaks make a huge difference. You’re telling the AI not just what to write, but how to write it and what to leave out.
But even if, after all this, the results are not perfect, you can always do some iterative prompting. Instead of trying to nail the perfect prompt on your first go, start simple and build from there. Think of it like layers in a cake; each prompt adds a new layer of clarity and precision until you’ve got exactly what you need. For example, say you start with a broad prompt like, "Explain renewable energy." The response will probably be vague and all over the place. So you refine it: "Focus on the advantages of wind energy compared to fossil fuels." Now you’re getting closer, but maybe it’s still not quite what you’re looking for. So you add another layer: "Rewrite the explanation for a 10-year-old audience using simple language and examples." And now the results hit the mark. Could you have written one detailed prompt up front? Sure! Something like, "Explain renewable energy, focusing on the advantages of wind energy compared to fossil fuels for a 10-year-old audience using simple language and examples." That might work for simple tasks, but as your requests get more complex, iterative prompting shines. It lets you refine and add detail step by step without overwhelming yourself or the AI.
Part of that specificity from earlier was giving the LLM information about style, tone of voice, formatting, that sort of thing. I’ll admit these little things don’t have a huge impact on general prompting, but if you need a specific format, style, or tone of voice, you should incorporate these things into your prompts. This can include formats like lists, tables, essays, or specific writing styles like poems, song verses, and so on. Take this vague prompt: "Tell me about the history of computers." It’s fine, but you’re leaving a lot up to chance. Instead, try: "Write a timeline of major events in computer history, formatted as a bullet list. Include five to seven key milestones with one sentence explaining each." Now AI knows exactly what you want: a timeline, a specific number of events, and concise explanations. Or if you’re going for something creative, like, "Write a Shakespearean sonnet about space exploration," the key is to name the format or style explicitly. Whether it’s essays, lists, tables, poems, or song lyrics, spelling it out ensures the response matches your expectations.
But sometimes, just telling the AI what you want isn’t enough. Instead, you show it by providing one or a few examples of what you’re looking for. You give the AI a template to work from. This approach is especially powerful for projects, for niche tasks, or creative projects where context or style matters a lot. For instance, say you want a chord progression in the style of the Beach Boys. A basic prompt might be, "Write a chord progression in the style of the Beach Boys." You might get something decent, but if you add an example, the results are next level. Try this: "Write a chord progression in the style of the Beach Boys. Here’s an example." With the example, the AI knows the vibe you’re aiming for. It’s like handing it a starting point to riff off of, ensuring the results are way closer to what you want.
Fused-shot prompting isn’t just for music; you can apply it to writing styles, journalistic, academic, humorous formatting, tables, outlines, scripts, specific tones, or perspectives. For example, "Write a dialogue between two people debating AI ethics." At this point, we already can write a perfect prompt that will give us exactly the results we want. However, there are still a few useful techniques to keep in mind, like the chain of thought thing. This technique is all about structuring your prompts like a checklist or guideline. It works wonders for more complex or multi-part requests, helping the AI think through the task logically and stay on track. Take this vague prompt: "Explain the pros and cons of renewable energy." Sure, it will give you something, but it might miss key points or wander off-topic. Instead, try this: "Explain the pros and cons of renewable energy by addressing the following: environmental impact, economic considerations, availability and scalability, long-term sustainability." Now you’ve mapped out exactly what needs to be covered. The AI follows your structure, hitting every key point without throwing in irrelevant fluff.
When you’re dealing with big problems and huge tasks, one of the smartest moves is to split them into smaller chunks. Not only does this make things more manageable, but it also dramatically reduces errors or hallucinations when the AI starts making stuff up. Complex tasks have a higher chance of going off the rails, so breaking them down keeps everything on track. Let’s look at this overloaded prompt: "Explain the causes, effects, and potential solutions for climate change." That’s asking a lot in one go. The AI might give a decent answer, but chances are it will miss key points or blend everything together in a messy way. Instead, break it into smaller prompts like this: "List the top three causes of climate change," "Describe the main effects of climate change on agriculture," "Suggest two practical solutions to combat climate change." Now, instead of one overly ambitious request, you’ve got three clear, focused tasks. AI will nail each one individually, and you can piece them together for a polished final result. And like before, this works great for simple tasks but shines even more for complex ones. The bigger your request, the more splitting helps.
And just in case you’re still not satisfied after all that, you can always ask your chosen LLM to help you with prompting. AI is great at breaking down and rephrasing tasks; it’s literally built for this. Paste your prompt and say, "Refine this prompt to make it clearer and more effective." For example, "Explain the causes, effects, and potential solutions for climate change." It might suggest splitting up or adding more specifics like tone, audience, or formatting. As a result, not only do you get a better-structured prompt, but you’re also learning how to improve your own prompting skills.
All right, let’s get real nerdy for a sec. If you’re into prompt engineering, there are these things called parameters—basically numbers you can tweak to control how an AI responds. Take temperature, for example; it’s like the randomness dial. Crank it up, and you will get super creative, chaotic responses. Turn it down, and you will get the AI equivalent of a straight-A student sticking to the rules. But every language model handles these parameters a little differently. In ChatGPT, Claude, or Gemini, you just type "temperature" followed by a number like 7 or 1.0. Easy, right? But switch to Mistral, and suddenly you’re using an equal sign instead of a "Y." Who knows? It’s just their thing. And that’s not all. You’ve got max tokens, which controls how long the AI will ramble on, top P for creative diversity, and top K, which is like saying, "Pick the best words, but not too many options." Messing with these lets you fine-tune your output like a pro.
Now, if we’re talking images, the game changes. Some principles carry over—good prompts equal good results—but the tools get picky. For example, DALL-E is your chill buddy; you can feed it a straightforward description, and boom, magic! But something like MidJourney? You better know the exact phrasing, like you’re talking to an art snob who only responds to just the right tone. On the surface, it works kind of like language models, but under the hood, it’s a somewhat different beast. Say you type something like, "A futuristic city at sunset," and the tool goes, "All right, I know what you’re saying," and starts turning that prompt into visuals. Here’s how your words get split into chunks, like "futuristic city" and "sunset." The AI has a number language where it knows how these ideas connect. It’s not just making wild guesses, though; it’s working off a massive database trained on billions of images with image descriptions. Think of it like showing an artist a zillion pictures and saying, "This is what futuristic could mean," or "This is what a city looks like." And here’s the cool part: it doesn’t copy; instead, it mashes together styles, vibes, and concepts it learned to cook up something entirely fresh.
So how does it actually make the image? It starts with random noise, like the static on an old TV, and gradually tweaks that mess until it matches your description. This magic process is called diffusion. The AI refines the noise bit by bit, turning chaos into something awesome. Now, when you’re prompting for image generation, the rules are a little different from text models. You’re not saying how to draw; you’re just describing exactly what you want to see. When it comes to crafting killer image prompts, the secret sauce is structure. You want clear, detailed instructions that give the AI everything it needs to blow your mind with the results. Here’s a super simple three-step framework to follow: 1. Subject, 2. Description, 3. Style/Aesthetic. Put it all together, and here’s what a solid prompt looks like: "The Batmobile stuck in Los Angeles traffic, impressionist painting, wide shot." Boom! Clear subject, vivid description, and an aesthetic that paints a perfect picture. Using this basic framework, you’ll not only get better results but also set yourself up for advanced techniques later. Trust me, once you start tweaking these elements, you will unlock next-level creativity.
And just in case you want more cool AI tools, visit aimaster.me, where we post detailed reviews of the coolest AI tools out there. Your subject is the foundation; think of it as the main character of your visual story. It’s usually a noun: dog, spaceship, guitar, waterfall. If you’re vague, like saying "joy" or "freedom," the AI is just going to shrug and throw something random at you. Stick with solid, tangible things. Want better results? Spice it up with adjectives. Don’t just say "cat"; say "a fluffy black cat with glowing green eyes." Already more vivid, right?
Next up, build the description. Add the context: What’s your subject doing? How are they doing it? What’s happening around them? Details matter. Saying "a dragon flying in the sky" works, but saying "a red dragon soaring through stormy clouds, lightning illuminating its scales as it breathes fire into the night" takes it to a whole new level. The background? Don’t skip it; it ties the whole vibe together. Finally, wrap it in style/aesthetic. This is like picking the filter for your image. How do you want it to look? Is it a photo, a painting, or maybe a 3D render? What’s the art style—impressionist, cyberpunk, or gothic vibes? Any artist influence? Want to scream "Picasso" or "Studio Ghibli"? Say it! How about framing? Do you want a close-up of the subject’s face or a dramatic wide shot?
And like I said earlier, there are rules to follow as well. The first rule of prompting: don’t overthink it. Just describe the image like you would to a friend who’s never seen it. But here’s the thing: lists of words can feel a little flat. Let me show you what I mean. Here’s a basic prompt: "Cat, urban street, cyberpunk, neon, nighttime, high quality." It’s clear enough, but it’s also kind of bare bones. Now imagine if instead of a list, you painted a whole scene: "A sleek black cat perched in the rain on an urban street in a glowing cyberpunk city at night. Neon signs in electric blues and purples reflect off the wet pavement, casting a dreamy glow. The cat’s cybernetic eyes shimmer softly as it watches hover cars zip through the misty air in the background." See the difference?
When it comes to prompt length, there’s no one-size-fits-all. Some AIs love the simplicity of short prompts; others thrive when you get into all the nitty-gritty details. You, as a prompt engineer, have to find the right prompt length depending on the complexity of the image you’re trying to generate. Check this out: this is a short prompt: "A snowy mountain range at sunrise, golden light, hidden icy peaks with a lone climber in the distance." Short and sweet, right? Gets the job done. If you want something simple, but sometimes adding a bit more description brings out the magic. Let’s expand on that: "A breathtaking snowy mountain range at sunrise, with golden light illuminating hidden icy peaks and a lone climber in the distance, creating a sense of adventure and tranquility." Now we’re starting to feel the scene. But what if we wanted to go all in and add every little detail? Here’s how that could look: "A breathtaking snowy mountain range at sunrise, with golden light illuminating hidden icy peaks and a lone climber in the distance, surrounded by a serene atmosphere and the sound of soft winds whispering through the trees." Now we’re painting a masterpiece!
Think of short prompts as quick sketches; they’re great for getting ideas down fast. Medium prompts are like rough drafts, balanced in creativity and control. And long prompts? Those are your detailed masterpieces, where you have the final say over every element. With practice, you’ll know exactly when to keep it short and when to let your creativity flow.
The best part is that every time you refine your prompts, you are sharpening your vision for the perfect image. And sometimes, you just have to be negative. You have to tell AI what you don’t want in your image. That’s called negative prompting. Think of it as a filter to cut out distractions and unwanted details. Whether you’re trying to avoid clashing styles, irrelevant objects, or specific colors, negative prompts help the AI zero in on your vision. In some image generators, there’s a specific area for a negative prompt where you just list stuff. In others, you need to modify the main prompt, but it’s really easy to do. You just type "avoid" or "exclude" and then list the things you don’t want.
Let’s say you’re creating a peaceful beach scene. You’re imagining crystal-clear water, white sand, and a serene vibe. But instead of tranquility, the AI gives you a modern beach resort with buildings and umbrellas everywhere. Here’s where negative prompting shines. Start by listing the elements you’d prefer to exclude. Be specific. Instead of just thinking, "I don’t want distractions," pinpoint exactly what those distractions are. Is it a particular style? A certain type of object? Once you’ve identified what to avoid, add those elements to your negative prompt. Keep it specific and direct, as vague terms might not work as well. So let’s say our main prompt looks like this: "A magical forest bathed in soft moonlight, glowing mushrooms scattered across the forest floor, and ancient trees rising toward the sky." Then the negative prompt would be: "Avoid cabins, pathways, buildings, fences, and artificial lighting."
One thing that often slips through the cracks when working with AI image generators is resolution and quality settings. These might seem like afterthoughts, but they can make a huge difference in the final look of your image, especially if you’re planning to print it or use it in larger projects. Most AI tools produce images at 72 DPI (dots per inch) by default. This is great for screens, websites, social media, or digital presentations, but 72 DPI isn’t ideal for printing or high-detail projects. If you try to print an AI-generated image at full size, you might notice some fuzziness or pixelation. And then there’s the size limit; many generators max out around 1,024 pixels in one dimension, which is fine for smaller projects but can feel restrictive for more demanding needs. That’s why it’s important to understand how resolution and layout fit into your prompt and strategy. Even if the AI has built-in size limits, you can nudge it to work better. Clarity and detail are key with your prompts. Mention terms like "high resolution," "4K," or even "detailed textures" to encourage the AI to pack more visual richness into the image, even if the resolution technically stays the same.
Layout is another area where your prompt can steer the AI. Use terms like "square," "landscape," or "portrait" to influence the image’s orientation. While you can’t always specify exact dimensions, suggesting an aspect ratio helps the AI craft images that fit your vision better. Each image generation platform has its quirks when it comes to resolution. Upscaling DALL-E is hands down one of the easiest tools for resolution tweaks. You can directly prompt it for different formats and even upscale the image later with minimal hassle. Need a larger version or a different layout? Just ask! MidJourney, while upscaling is also possible here, requires a bit more experimentation. You’ll need to figure out the right settings and commands, but once you get the hang of it, the results can be stunning. Pro tip: using terms like "high detail" or "photo realistic" can help nudge MidJourney to generate sharper outputs from the start.
Oh, and speaking of MidJourney, we just put together a full guide to it. Make sure to check it out; it’s packed with tips to take your image generation to the next level. At its core, all prompting is about giving clear instructions. AI is like a kid who doesn’t know what to do but has all the tools for it. So if that’s an LLM, tell it exactly how to write. If that’s an image generator, describe, describe, describe. Simple as that. Thanks for watching, and see you in the next video!