Brief Summary
This video provides a step-by-step guide on how to use ChatGPT to create stunning and realistic AI images, even without prior experience. It covers essential aspects such as crafting effective prompts, refining image styles, using photography terms, creating realistic images of people, and addressing ethical considerations.
- Crafting effective prompts using a specific formula.
- Refining image styles by specifying art styles or eras.
- Using photography terms to control lighting and depth of field.
- Creating realistic images of people by detailing their characteristics and emotions.
- Addressing ethical considerations to avoid deception, harm, or copyright infringement.
The video introduces a guide for writers, marketers, communications professionals, and content creators on using Chat GPT to generate realistic AI images. The creator, Aga, who runs an online education business, promises to share use cases, examples, and less obvious tips. She encourages viewers to subscribe for more content on AI, content creation, and productivity.
To generate an image in Chat GPT, you need to enter a prompt, which is a description of the desired image. A basic formula for an effective prompt includes the subject, scene or setting, style, and details. Using a detailed prompt results in a more specific and less random image compared to a basic prompt. For example, a basic prompt like "create an image of coffee" yields a random image, while a detailed prompt like "a steaming cup of coffee on a rustic wooden table, morning sunlight coming through a window in soft cinematic lighting" provides a more specific and visually appealing result.
When creating an image for a specific piece of content, you can specify the dimensions in your prompt. For example, you can request a square image for Instagram or a wide image for a blog header. By including format details in the prompt, you can avoid additional editing after the image is generated. For instance, prompting "a square image of a vibrant colorful smoothie bowl with fruit and seeds, tropical vibes styled like a trendy Instagram post" will generate an image with the specified format and style. If you're not satisfied with the result, you can ask Chat GPT to create several alternatives to save time.
You can refine the style of an image by specifying the art style or era in your prompt. Using the same prompt for a smoothie bowl, you can generate different images by specifying styles such as a 1970s magazine ad or a children's book illustration. This allows you to create images that align with your brand and desired aesthetic. Specifying the style and era helps Chat GPT produce diverse images, providing you with more options to choose from.
Using basic photography terms in your prompts can help create a specific feel for your images. Terms like "shallow depth of field" can create a blurry background, while "backlit" or "golden hour" can specify the type of lighting. For a more polished look, you can use "soft studio lighting." For example, a prompt like "a woman reading on a couch in natural window light, shallow depth of field, candid photo style" will generate an image with natural lighting and a blurry background, resembling a candid photo. Familiarizing yourself with photography terminology ensures you get the desired look and feel in your AI-generated images.
When creating images of people, it's better to create a character with specific details rather than just asking for a picture of a person. Adding parameters to your prompt can significantly change the outcome. For example, instead of just saying "a woman standing outside," you can specify "a 30s something South Asian woman with curly hair wearing casual jeans and a white linen blouse standing outside a small bookshop smiling naturally." This provides more detail about the character's appearance, clothing, and environment, resulting in a less random and more specific image. Grounding the character in a real environment, such as a local market or a messy living room, can also make the image more realistic.
AI-generated images tend to be perfect, lacking imperfections that make them look fake. To make images more realistic, introduce imperfections in your prompts. For example, ask for "dirty old trainers" or "a cafe with slightly messy tables in the background." Real life is messy, and your audience is more likely to relate to images that reflect this. Similarly, AI often generates images of conventionally good-looking people. Modify your prompt to request someone who is not conventionally good-looking or specify features like thinning hair or uneven skin. Anchoring images in culture or time, such as "a cozy autumn street in Brooklyn with brownstone buildings" or "a millennial woman working on a laptop in a cafe 2020 style," can also enhance realism and familiarity.
Before creating images with Chat GPT, consider the ethical implications. Avoid using AI to deceive your audience or create harmful, misleading, or copyright-infringing images. Be transparent about using AI-generated images, especially in a commercial context. Ethical considerations are crucial to ensure responsible and honest use of AI in image creation.
Summarize any YouTube video in seconds
Get summaries like this one for any video, in your language, with the free Summ app.
Transcript
Full transcript of “How to Make Images with ChatGPT for Beginners”. Click a timestamp to open the video at that moment. Need another video? Use the YouTube transcript generator.
0:00 if you're a writer marketer communications professional or content creator and you've ever struggled to find the perfect image before this video is for you i'm going to show you step by step how to use Chat GPT to create stunning and realistic AI images even if you've never done it before i'm going to include some use cases and lots of examples and I will share some less obvious tips as well if you're new here I'm Aga i run an online education business with more than 1.5 million followers across my social media platforms on this channel I share my top tips on content AI and productivity so hit the subscribe button and let's get started
0:47 so if you want to make an image in Chad GPT you just need to go to the Chad GPT website so you can generate an image by entering a prompt into this box a prompt is just a description of the image that you want to make so what do you actually want to see here's a basic formula to include in your prompt subject scene or setting style and details so let's compare what it looks like when you use a very basic prompt versus when you use the formula I've just shown you so create an image of coffee it takes some time to create those images so make sure you're patient but I'm going to enter a slightly different prompt using the formula I've just given you so a steaming cup of
1:32 coffee on a rustic wooden table morning sunlight coming through a window in soft cinematic lighting so I've included lots of details about the vibe that I want to include in the image where this coffee is what it looks like the temperature of the coffee and also some details about the lighting that I'm after so you can see there is a difference and the main difference is that the first image is very random so if you're looking for something random and you don't mind what you get that's fine but if you have a specific idea about the type of image that you need make sure that you include as much detail as possible in the prompt one thing to mention here is that because you're likely creating an image for a specific piece of content you can specify the
2:20 dimensions in your prompt you can tell Chad GPT that you need a square image for Instagram or a wide image for a blog header so I'm going to add this prompt into Chad GPT a square image of a vibrant colorful smoothie bowl with fruit and seeds tropical vibes styled like a trendy Instagram post so I've included information about the format so that I don't actually have to do anything with the image afterwards and this is what it comes up with so it's got all the details that I've asked for if I'm not happy with it I can always do it again or I can ask Chad GPT to create several different alternatives that is something that can really save you time as well so in your prompt you can just say "Give me five alternatives of this picture." The next thing you're going to
3:08 do is you're going to refine the style so first you can specify the art style or the era for the image so we're going to use the same prompt that we used for the smoothie bowl but we're going to do it in two different styles the first style is going to be a 1970s magazine ad style and the second one is going to be children's book illustration style so this is just to show you the differences between what Chad GPT comes up with when using different prompts about style and the era that you want the picture to resemble so you can see they are very different images so now I've got lots of different things to choose from and I just want to make sure that whatever I've created aligns with my brand and just doesn't look out of place you can also use some basic photography
3:58 terms to create a certain feel like shallow depth of field for a blurry background include back lit or golden hour in your prompt for a specific type of lighting or you can say soft studio lighting if you want a more polished professional look let's see what it does with a prompt about lighting so a woman reading on a couch in natural window light shallow depth of field candid photo style so this is great that the lighting is natural there is a blurry background and it does look quite candid so you can familiarize yourself with some basic photography terminology to make sure that you get the right kind of thing that you're looking for the next thing I want to talk about is images of people so when you're creating an image of a person it's a good idea to create a character
4:49 rather than just saying that you need a picture of a person or a man so let me show you what I mean so let's see how a picture can change when you add different parameters to your prompt so I'm going to just say a woman standing outside which is obviously going to give me quite a random picture and I haven't specified anything about the woman or where she is as you can see the picture is quite random it might work for you if that's what you're looking for but um it's not what I'm looking for so I'm going to specify that I would like a 30s something South Asian woman with curly hair wearing casual jeans and a white linen blouse standing outside a small bookshop smiling naturally so I've given quite a bit of details about what I want her to look
5:35 like what I want her to wear and also where exactly she is and how she's behaving so you can see that the picture really reflects uh the description the prompt that I've given uh it is less random obviously to save time I can ask Chad GPT to create five alternatives of this image so I've got something to choose from it's also a good idea to ground the character in a real environment to make it look more realistic for example at a local market in a messy living room on a rainy pavement my next tip is important especially when creating images of groups of people so for example you might be creating an image of a group of people working together for a blog post if
6:22 you're not specific about the differences between the people in the picture Chad GPT may give you a group of people that look very similar to each other so to introduce a little bit of variety you can either describe each person or you can just give Chad GPG a general prompt to say that you want a variety of different ethnicities ages and looks to make your image even more realistic you can describe the emotion so instead of just saying a person looking at a laptop I'm going to introduce some emotion into the prompt so I'm going to say a young man looking at a laptop with a surprised and excited expression as if he just as if he's just received good news so this has
7:08 given me a very different image an image where you can actually see what this man might be feeling sometimes you might need an image where the person isn't actually visible so in your prompt you can say seen from behind or you may need a picture of hands doing an action so you can just say a person's hand holding an iPad my next tip is make it more realistic so AI generated images tend to be quite perfect there's not many imperfections in them and sometimes you just need to use the right prompt to introduce some imperfections because perfection just looks fake real life is messy but there's nothing wrong with that that's just how life is so for example you can ask it for dirty
7:58 old trainers or a cafe with slightly messy tables in the background nothing is perfect in real life and your audience is likely to relate more to a picture that shows what life is really like the same thing applies to people so Chant GBT and AI in general tends to generate images of people who are conventionally good-looking so what you can do is you can modify your prompt to tell Chad GPT that you want somebody who is not conventionally good-looking or you can talk about the specific features that you want changed like you may want thinning hair or uneven skin sometimes perfect people are fine for your brand this is this might be what you need but often people relate more to
8:46 images of people that look more like an average person that you see out and about you can also anchor your images in culture or in time you can create images that will be familiar to your audience like a cozy autumn street in Brooklyn with brownstone buildings or a millennial woman working on a laptop in a cafe 2020 style and finally ethical considerations so before you go and create a bunch of images with chat GPT it's important you think about ethics so make sure that you're not using AI to deceive your audience or to create harmful or misleading images or images that infringe on someone else's copyright and finally be transparent about using AI generated images
9:33 especially if you're using them in a commercial context if this video helped you make sure you subscribe to my channel and give this video a like i really appreciate it and I'll see you next time
