Goodbye Midjourney... Ultimate ChatGPT-4 Image Guide 2025 (ChatGPT for beginners)

Chad GBT 4. 1 will soon replace the 4. And it will have the best image generation you'll ever see. More features, more controls, more of everything, and it will surely be heavily based on that new image generation, the 4. And you need to know exactly how to use it when it launches. In this video, I will show you exactly how to prompt it, what it can and can't do, why it outshines my journey, and I will even give you real life examples of how we use

it. To access the new image generation, you don't need to do much. First, go to the website and sign in. Then, brace yourself. You need a $20 subscription. Yes, they did roll out a free version, but it only lets you generate free images a day, just like Dolly used to. That's barely enough to dip your toe in, let alone follow a full tutorial. So, if you actually want to learn and use it, you will need that subscription. Plus, users still have a few limits. Five images per hour. And if you are a pro subscriber, you can generate away with no real cap for now. Just keep in mind, none of this is set in stone. So, things might have changed by the time you're watching this. But once you're in, all you have to do is start a new chat and prompt away. All right. So before we dive in to

actually write in prompt, there is something you you've got to know. It's not dolly. It's not an add-on, not a side feature. It's a whole new system built directly inside GBT40. This means can turn out photorealistic images, edit and transform any pics you give it and follow crazy detailed instructions, including putting text on the image exactly where you want it. because it's tucked so deeply into the 40's architecture. Can use all of GBT's knowledge to produce images that aren't just pretty, but actually useful and precise. You can fire it up by clicking the create image button or just type something like generate me or draw or create an image of any phrase that sounds like you're asking for a picture. Also, from now on, Chad GBT will default to this new algorithm. So, if you still want the all dolly, you can find it off to the left in a special menu. I like my image generation like my coffee strong and refreshing. I use it every day. I just love chat GBT even though it lacks a crucial feature for me transcribing notes. And I would really prefer a better tool for transcribing my voice notes like Letterly. Letterly is one of those AI apps that feels like it's been tailorade to simplify everyday tasks. I start by speaking naturally, just let my ideas flow without worrying about structure or grammar. Then the magic happens. Letterly AI instantly transforms your spoken words into clear, polished text I can use anywhere. Emails, social media posts, brainstorming notes, or even professional documents. What really stands out is how flexible it is. I can choose from over 25 rewrite options to match my style. Whether I want a friendly tone for casual chat or a more formal approach for business communication, bullet points for a quick list. Here you go. Short, snappy caption for Instagram. Easy. Literally handles everything from minor tweaks to more significant edits so I can say what I need without worrying about the final form. Later supports over 90 languages so it's useful for bilingual people or international teams. You also get handy functions like screen off recording offline mode and instant syncing across devices. And you've got a tool that's easy to weave into any workflow at any moment. For anyone who's felt that spark of inspiration but missed the chance to jot it down, Latterly captures ideas when they're fresh and converts them into usable text in seconds. Sometimes all it takes is a quick nudge from the right tool to bring your ideas to life. Laterally is that nudge turning everyday thoughts into well-crafted text so you can focus on what really matters, sharing those insights with the world. And right now, AppSumo is having an AI week offering a lifetime deal on Letterly. You should check it out. Pay once and you're set for good with no monthly fees around the corner. The link is in the

description. The biggest rule is get descriptive. Really, really descriptive. With Dolly, we kept adding detail because it would forget or skip stuff. Now, we're still piling on the details, but because can actually juggle all of them. You can describe the overall scene, the little objects in it, the people, the lighting, you name it, up to about 20 separate elements. After that, it might start acting a bit confused. By this, I mean something like 10 different people or 10 different objects, each with its own mini description. For example, I can ask for a shot with 10 different people in it, each with different clothes and poses, and it nailed it. Or you might try a big scene with animals, weather effects, plants, people. Anna can still fit them all in. Natural language is key. No need for complicated prompt formulas. Just be direct and clear. If you've got multiple people, give them names instead of he or she so the AI knows who's who. And use short, simple sentences to keep everything easy to

follow. If you've just generated an image and want to make some changes, here's how you do it. First, click on the image to open it full screen, then look for a select button. Go ahead and click that. Highlight the area you want to tweak and in the prompt window, explain exactly what edits you'd like. One heads up, there's no way to change the brush size. So try not to highlight more than you need. If you make a mistake, you can use the undo and redo buttons at the top or hit cancel to scrap the selection and start over. Do keep in mind that the edits might go a little beyond the area you highlighted, which is normal for most AI image editors. They all share similar quirks. So most of the advice you've seen for Dolly also works here. In fact, we've got a Dolly guide you can check out, and every tip there applies to the new image generator. Over on Geek Academy, we have even more guides like that, plus PDFs you can actually save for later, blog posts with prompt and tips and curated tool collections for AI fans. There's also a built-in image generator you can play with, super easy to use, and a handy tool finder that helps you pick the perfect AI tool for any job or even answer random AI related questions you've got. We also offer voice chat with AI Master if you prefer talking to a digital copy of me. And on top of all that, we just launched our new educational course on generative AI. Seriously, if you're into AI, Geek Academy is your new best friend. Check the link in the description and come join us. By the way, you don't have to use the selection tool to make changes. You can just type in your desired edit at the bottom of the editor interface. And if it's area specific, either explain that in your prompt or highlight it with the brush. Once you're happy with the final look, just hit download. This works exactly the same on the mobile app, too. Same algorithm, same

workflow. One of the coolest things is how well it can handle and understand reference images. If you have a photo you want to recreate, like a stock image with a watermark, you can upload it and GPT40 will create something similar, but different enough that you're not stepping on any copyright toes. You can also pick a shot you love for its framing, layout, or mood, and use it as a base idea. If you feed it multiple shots of

the same character, maybe you, your favorite video game hero, or even your pet hamster, it can keep that character's look consistent across every new image you generate. Older models usually mess this up, but 4 is surprisingly reliable here. The only catch is that all your reference images need to show the same character in the same style. So, if you want a digital clone of yourself, then stick to photos of your face. If you want your game character's face, then grab some screenshots. This works for full body shots, too. But try to use images where the outfit is the same in each photo. And unlike other image tools, there's no waiting around for the AI to train on your character. Once you've uploaded your images, you can start creating right away. Just remember, the more references you give it, the better your final images will

look. Because Chad GBT is so good with uploaded images, you can get really creative. For example, maybe you've sketched a rough drawing on a napkin. You can upload it and tell it to turn it into a polished image. It usually does a fantastic job, especially if your prompt is detailed and clear, so it doesn't have to guess anything. If you were an architect who wants quick, decent mock-ups, you might want to know that the default output resolution is kind of low. So, if you need something more high- res, be prepared to ask ChatGpt to upscale it or use a separate app to do that. It's still a lot faster than doing everything by hand

though. Another superpower is text generation. AI used to be hilariously bad at drawing words in pictures, but now it is light years ahead. And because it's built right into GBT's core, can generate text all on its own. This makes creating things like comics, menus, or invitations super easy since you can prompt it for both the image and the text together. There are still some limitations here and there, but we'll chat about those soon. If you need a bigger block of text in your image, just paste it straight into your prompt. It might feel old school, but it gives you total control over the final result, especially if you're working in a language that doesn't use the Latin alphabet. And if you need the text to show up on something like a t-shirt or a sign, Chad GBT can handle it. Warping the letters so they match the folds or the

the angle. If you watched our earlier video about Chad GBT and Dolly, you will remember that iterative editing was still in its infancy back then. But now with Chad GBT's improved memory, iterative image generation has become really practical. What does that mean in plain English? If you're not totally sure about your initial prompt, you can split the whole process into multiple steps. Start with the basic layout. Decide on the composition and the main subjects. Then gradually add any missing elements or fine-tune existing ones. Once the layout is set, move on to colors and textures. After you're happy with that, you can tweak the style, lighting, and any other finishing touches you want. Of course, you can also pack all these details into one huge prompt and still be able to edit later if something's off. But the step-by-step approach is really the star of the show here, letting you refine everything at your own pace. The improved text algorithms go

even further when it comes to making comics. And yes, you can actually generate full-fledged comics now. The results are surprisingly decent. The characters look consistent across different panels. Each panel has its own mini story line, and the comic can feel like it was made by a person. For instance, if I ask for a four panel comic about one dog reacting to another dog, I'll actually get four cohesive panels that form a short readable story. Chad GBT can also generate simple infographics, and it won't create insanely detailed charts, but diagrams like a basic water cycle or a hierarchy of needs chart are well within its capabilities. The prompt can be as vague or as specific as you want. If you're feeling creative, give it a general idea and let it fill in the details. If you want more control, list out each element you need. It all depends on how much you trust Chat GPT to do the heavy

lifting. Everything we've covered so far is already impressive, but here's something that's even better. Real image editing on photos you upload. We've touched on uploading images for reference, but you can also directly fix or modify them. For example, if you have a picture with distracting reflections on a glass window, just upload it and ask it to remove those reflections. In a few seconds, you get a cleaner image. Sure, it may look slightly more AI generated afterward, but throw on a filter and nobody on Instagram will even notice. And you can do much more than remove reflections. Maybe you'd like to change your shirt color, remove glasses, or swap hairstyles in a selfie. ChatGpt will focus on precisely those features without messing up the rest of your photo. It's super handy and almost too easy. You can also ask it to do things like color grading, adjusting the time of day, or even changing the weather in the picture. I can take a regular daytime street picture, tell Chad GPT to make it night, and boom, it's suddenly a nighttime shot with the right levels of contrast and brightness. Then I can say, "Turn this scene into golden hour and add rain. " And watch it transform yet again. It doesn't just slap a filter on top. It actually changes how the photo looks in a realistic way. Honestly, it's one of the best new additions I've seen in Chad GBT. Although, I'm still hoping they'll improve the final output resolution soon. And don't forget that you can add new stuff to your images. Just select an area and ask Chad GBT to put something there, like a person walking their dog. The AI blends it in so nicely, it might look like it was there all along. It's basically Photoshop for anyone who doesn't want to learn complicated tools or pay for subscriptions. Just upload your image, explain what you want, and ChatGBT does the

the rest. This next one is pretty straightforward. You can now create images with a transparent background, which is a total bonker for making stickers or removing backgrounds. You just ask Chad GBT for an image with a transparent background and you've got it. No extra steps needed. And you can now download the image directly from the chat by clicking here. No need to ask for a separate

link. Let's say there is an image you already like, maybe something you found online, drew yourself, or took a quick snapshot of. Now you want that same image but in a totally different style. Just upload it to Chad GPT and ask for a redraw in your chosen style. For example, you could go for the popular Giblly look with soft pastel colors and delicate line work. Chat GPT preserves the basic framing, positioning, facial features, and background elements while still transforming the image to match the new style. And it doesn't have to be just photos or Gibli art. Any style you can imagine is fair game. If you can dream it, Chad GBT can draw it. The best part is that it actually

contrast and brightness. Then I can say, "Turn this scene into golden hour and add rain. " And watch it transform yet again. It doesn't just slap a filter on top. It actually changes how the photo looks in a realistic way. Honestly, it's one of the best new additions I've seen in Chad GBT. Although, I'm still hoping they'll improve the final output resolution soon. And don't forget that you can add new stuff to your images. Just select an area and ask Chad GBT to put something there, like a person walking their dog. The AI blends it in so nicely, it might look like it was there all along. It's basically Photoshop for anyone who doesn't want to learn complicated tools or pay for subscriptions. Just upload your image, explain what you want, and ChatGBT does the rest. This next one is pretty straightforward. You can now create images with a transparent background, which is a total bonker for making stickers or removing backgrounds. You just ask Chad GBT for an image with a transparent background and you've got it. No extra steps needed. And you can now download the image directly from the chat by clicking here. No need to ask for a separate link. Let's say there is an image you already like, maybe something you found online, drew yourself, or took a quick snapshot of. Now you want that same image but in a totally different style. Just upload it to Chad GPT and ask for a redraw in your chosen style. For example, you could go for the popular Giblly look with soft pastel colors and delicate line work. Chat GPT preserves the basic framing, positioning, facial features, and background elements while still transforming the image to match the new style. And it doesn't have to be just photos or Gibli art. Any style you can imagine is fair game. If you can dream it, Chad GBT can draw it. The best part is that it actually remembers your past images, meaning you can generate several pictures in a row, then cross reference them. For instance, you could create a portrait of one person, someone else, and then ask it to merge those two portraits into one image. However, this enhanced memory sometimes glitches, causing every new result to look too much like the previous one, especially if you don't modify your prompt. In the past, you might have changed a few words and gotten a totally different picture, but now you might end up with nearly the same shot just with a few tweaks compared to the previous version. This issue isn't very common, but it's something you should be aware of. If it happens, you can start a fresh chat or rewrite your prompt to make sure it doesn't look like your old one. Let's say you took a photo, but

your shaky hands made it blurry. Upload that photo to Chat GBT. Then ask it to sharpen the image, make it crisp, and add details. The result won't be a perfect replica of your original picture. Some features might not be spot-on, but it's still miles better than a blurry shot. In many cases, the AI's adjustments turn out surprisingly well, so it's worth a try. You can also ask it to upscale your images and fill in missing details. I'd suggest first asking for extra detail so it packs in as much info as possible and then having it increase the resolution. Unfortunately, you can't directly set an exact resolution before downloading, but you can mention your desired resolution and then ask for a download link. Technically, scaling up is somewhat similar to just stretching the image, but Chad GPT does add a bit of anti-aliasing and smoothing, so the result looks reasonably polished. Definitely better than the original. Besides just adjusting resolution, you can also set a custom aspect ratio for your image. It's super easy. Simply write 16-9 or square in your prompt. You can even try panorama vertical 189 and so on. It's a quick and easy little trick for getting the exact shape you

want. If you want a truly photorealistic look, simply adding photo realistic to your prompt won't cut it anymore, and you'll still end up with images that look a bit AI generated. A smarter approach is to specify an actual camera and lens setup. Just pick one camera model and one lens type and stick with them during your session. If you prefer, you can even ask it to suggest a lens for each image and pair that with random camera model, giving you results that look way more realistic. Obviously, all the other tips from our Chad GBT and Dolly guide still work. So, if you're itching for more, go ahead and watch that video or download that PDF guide to image generation from Geek

Academy. While 4's image generation is impressive, it does have a few quirks. For instance, if you want a very tall or complex image, like a full poster or a page layout, 40 might crop it in ways you don't intend. It often snips off the bottom or edges of long images. And it can also trim text along the borders. One workaround is to mention extra space in your prompt or explicitly ask to expand the image so the model knows not to cut anything off. Occasionally, it adds random elements into your image. This usually happens if your prompt is too short and the model has to guess what goes in the scene. A good rule of thumb is to be really clear about what you do and don't want. If something weird shows up, you can say remove that and regenerate. GPT40 is better than older versions and sticking to your description, but small surprises can still slip in. If you pack too many distinct ideas into a single prompt, it might drop or mix them up. OpenAI says after about 10 to 20 separate concepts in one image, the model starts struggling. For example, a prompt asking for 30 different flags on 30 different animals is likely to produce errors. Some flags or animals might be off. Or if you request a super busy scene, like a crowded, where's Waldo style image with dozens of unique characters, don't expect every last detail to be perfect. Another area that needs improvement is rendering non-English text. It generally nails English letters and numbers, but it can stumble over scripts like Chinese, Japanese, Arabic, or Hindi. Sometimes you'll see characters that look halfway right, but are actually gibberish. The model isn't as strong at mapping those scripts correctly, especially if you want a long string of text. If you need accurate non-Laddin text, a good workound is to generate the image without text and add the lettering yourself later.

So, how do I personally use this improved image generation? If you've watched our older videos, you know we rely on AI to create thumbnails. And this time is no exception. But our approach has changed. Instead of giving ChadBt a basic prompt and hoping for okay results, I can feed it a photo of myself or anyone else and have ChadBt place that person in any setting I want. Take this thumbnail for example from one of our recent videos. It's a simple layout, barely any elements, no text. Yet, making it by hand would have taken hours thanks to all the 3D effects, blooms, light streaks, and so on. Instead, I asked Chad GBT to generate an image of the Tesla logo, destroying the Chad GBT logo, requested a dark background in blue, and then took a photo of Elon Musk and asked ChadBt to put him on the right side, and boom, thumbnail done. We don't always let ChadBt do 100% of the thumbnail work, though. Sometimes we just use it for individual elements. For instance, here's another thumbnail for one of our recent videos. The logo of this Chinese app was entirely made by Chad GBT along with a few other logos we didn't end up using. The biggest takeaway is that you shouldn't rely on Chad GBT to create text for YouTube thumbnails. Even if you specify the font style and show examples, Chad GBT often messes up text size and positioning. One workaround we sometimes try is feeding Chad GBT a bunch of our old thumbnails so it has a reference for how our text and layouts usually look. But even with the models improvements, we still like to touch up the final thumbnail ourselves. Each new version Chad GBT generates can randomly tweak parts of the image you wanted to keep the same. You might end up with a perfect look on one occasion, but something weird on the other. This goes double for faces. Be careful when you're editing faces you upload. For one iteration, the face might still be pretty close to the original, but by the second can get really wonky. This new image generation feature is definitely

fantastic. It can do so much more now, and it really opens up new creative possibilities. It's not flawless yet, but it's pretty close. And with every update, it just gets better. Who knows, maybe it will replace Photoshop or similar tools someday. For now, though, unless you're in a Plus subscription, using it to its fullest might be tough. So, feel free to jump into Geek Academy and give her image generator a spin. Thanks for watching and we'll see you in the next video.