* This blog post is a summary of this video.

Leveraging ChatGPT and DALL-E 3 to Create Powerful AI-Generated Images

Author: KI-PflaumeTime: 2024-02-11 19:50:01

Table of Contents

Introduction to Using ChatGPT and DALL-E 3 Together

Hi friends, I have to revise my conclusion from the last DALL-E 3 video. Combined with ChatGPT, the tool is extremely powerful. For many applications it is currently even stronger than Midjourney. Let's start at the beginning.

As a ChatGPT Plus member you can now also have images created with the new DALL-E 3 image generator. Since OpenAI owns both ChatGPT and DALL-E 3, I think they have retained the best features for their platform. And here we see the functions.

Core Benefits of Using ChatGPT and DALL-E 3 Together

Image generation from text. DALL E can generate images based on text descriptions. The text can be any as long as it is within the guidelines. Image resolution. The desired resolution of the image can be selected from three options. Wide, square or tall. That's what I criticized in the last video and yes, I'm even more pleased that it's now possible here. Image description prompts. Up to four different text descriptions, prompts, can be provided for image generation to obtain a variety of images from different perspectives or interpretations. Here you basically just have to give ChatGPT a simple prompt and it converts it into four more complex prompts and if you haven't specified an exact image type, it takes four different ones that are interesting for that topic. It's this strong interaction between chatbot and image generator.

How to Generate AI Images in ChatGPT

Of course, I’ll show you examples right away. The guidelines are definitely stricter than Midjourney. You can't generate politicians or public figures here. This is probably also related to all the lawsuits that OpenAI is currently having to deal with. Yes, and the topic of promoting diversity is definitely noticeable too. When it comes to people, for example, people from different backgrounds are always generated and accurately.

Entering Text Prompts

Here is one of my tests. I entered the image type here, photo. You have to use this if you want a photorealistic image. About shopping basketballs in a shop. And then at the back there is the aspect ratio, i.e. the aspect ratio 16 to 9. It's not quite 16 to 9, but he understands what I mean. After that I said, give me the JSON Codes of the image after generating. I did that because otherwise you can't get the seed here.

Specifying Image Resolution and Aspect Ratio

Yes, and here you can see straight away that I have seed 5000 everywhere. And that was the case throughout. That's why I'm interested, please test it and tell me if it works for you too 5000 was set as a seed. I don't know now whether the account is tied to this number. That would definitely be super interesting. Yes, or the system is not implemented yet.

Leveraging Multiple Image Description Prompts

I then simply took this first prompt here and inserted it again. Then the four images were generated here again. They're also very nice, I have to say. Except for a few heads here, but in principle it's going in the right direction. Now, of course, there are a lot of us, but that doesn't matter now. We can now say afterwards, give me the metadata for images 1 to 3. Exactly and then you will also get your prompt here. If you didn't enter that directly into the prompt, you can also see Seed 5000 here. I then simply inserted the JSON code here and the images were actually created, but you can see it's in a square format. This is because the aspect ratios were not included here. I didn't test it, so we can do it like this now. Including aspect ratios.

Guidelines and Limitations of DALL-E 3

Now here it gives us the aspect ratio, but of course this is incorrect code here. Then we just enter Size. Then he gives us the correct measurements here. Let's assume that this is our character here, then we would copy this here and re-enter this here. Simply insert the prompt or JSON code here and then you can read it directly. And this is what the image creation process looks like now. In that case, of course, he takes our prompts that we have given. Otherwise it will create four variants here. In this case, he now only takes three pictures because we only specified three. Image 1 to 3. And now the images would have to be spit out exactly as they were generated.

Avoiding Sensitive Topics and Public Figures

Yes, and these are the pictures. Exactly. And here he also takes the aspect ratio into account. So you have to write the size in separately again, otherwise it will only give you seed and prompt.

Promoting Diversity

I've already had a few examples generated. Here for my blog, for example. The girl who is putting on her handball shoes is a very good picture in my opinion . Directly in store. I definitely couldn't manage that with Midjourney. Exactly, and these images are great to include.

Using Seeds for Identical Image Recreation

I've had good experiences here by simply entering 16 to 9, i.e. aspect ratio, and then Picture About XY, i.e. a topic, and then it gives you these more complex prompts.

Saving Seeds to Reuse Characters and Objects

Yes, in my opinion that is a very good image that you can use. I will use that too. And exactly, here are a few examples. Yes, here, of course something like that is generated, but as I said, trial and error. Just try it out a few times. There are a few pictures that you can then use. It's not bad here either. But here, Respect the Sandbag, is of course pretty creepy. This could also be a sandbag cult.

Real-World Examples and Use Cases

Well, you definitely know what I mean. 16 to 9, Picture About the Price of Boxing Gloves. I find it here too. This is also an image that you can take. I've already used that too. Simply that it represents the price again. Or here, an online shop like this. Then you have to adjust the font.

Blog Articles and Social Media Posts

Exactly, I've already figured out a trick on how to do this with Canva. I'll show that in another video. But the results are good. Or here, different types of boxing gloves. Yes, good pictures come out of it. You can use them. Or here, the importance of boxing gloves. In the background there are a few injuries, injured hands. Yes, and of course that symbolizes the importance again.

Illustrations and Photo Collages

Well I have to say, the combination with the chatbot and the image generator ensures good results. Exactly, and if we now say we want to have several variations of image 1, then we simply enter. You can do this in German or English. I always enter the prompt in English, but in principle it also works very well in German, I have to say.

Conclusion and Summary

Yes, and then you get four more variations here. Yes, and here you can also see what I meant about the topic of diversity. So, in order to be able to create images at all, you of course have to select DALL-E 3 up here. So, now let's test the whole thing with our fictitious prompt. A world where all animals can fly.

Exactly, here we have an oil painting, an illustration and two photos. The whole thing always takes a while. So it is not possible to generate images super quickly here. This is definitely a negative point, but I think this will be further fixed in the future.

Yes, and we have really nice pictures here too. Exactly, I think it's sometimes even better than with Journey. Especially here the illustrations are very, very strong. So, to apply the whole thing, we'll just take the boxing gloves here. You can really have everything created depending on the situation and that's the good thing.

You can create beautiful images for subheadings here. That loosens up the whole article and yes, the results are simply beautiful. Another little trick is that if you click on download here, the image is usually two to four megabytes in size. But if you click on the image here and right-click to download the whole thing , you'll get it compressed.

Yes, strangely enough now also as a WebP file, but you have about a tenth of the size and if you want to upload it, that's very suitable. Yes, so my conclusion is, ChatGPT makes me spend another $20 every month. DALL-E 3 in ChatGPT is super, super strong and I'm just excited to see how it will improve in the future.

The results are now very easy to work with and yes, all of the images can be used . So it's a very, very good feature and yes, I will definitely use it for myself. Feel free to write your opinion in the comments. I would like to thank you for watching and if you would like to exchange ideas about the topic, please come to the Artificial Intelligence Germany Discord. You can find the invitation link in the video description. Until next time, all the best!


Q: How do I generate images in ChatGPT?
A: You can generate images by selecting the DALL-E 3 tab in ChatGPT, entering a text prompt, and clicking 'Create'. You can specify options like image size and aspect ratio.

Q: Can I recreate the same image multiple times?
A: Yes, using seeds. When you generate an image, save the prompt and seed. Re-entering them together will recreate the same image.

Q: Are there any content limitations?
A: Yes, you cannot generate images of public figures or sensitive topics that violate OpenAI's content policy.

Q: What file formats do the images use?
A: Downloading from ChatGPT uses JPEG/PNG, but you can right click and save as WebP for smaller file sizes.

Q: Can I use these images commercially?
A: You have broad commercial usage rights, but cannot claim the AI as your own creation or imply OpenAI endorsed the content.

Q: How are the images optimized for diversity?
A: DALL-E 3 aims to represent people inclusively with regards to race, ethnicity, gender, age, and other factors in generated images.

Q: Are the images high resolution?
A: Yes, up to 1024x1024 pixels currently. Higher than other AI image generators.

Q: What types of images can be generated?
A: All kinds - illustrations, landscapes, characters, objects, logos, and more. Both realistic and abstract styles.

Q: Can I use these for a commercial website?
A: Yes, they have broad usage rights for commercial applications, but you cannot claim AI generated content as your own creative work.

Q: Is there a limit on how many images I can generate?
A: Yes, based on your ChatGPT subscription tier. Paid plans get more monthly image generation quota.