While more attention is often focused on text-based AI offerings, AI image generators such as Dall-E, Stable Diffusion, Midjourney and others have come a long way.
Here is an overview of important offerings:
Dall-E 3
Like ChatGPT, Dall-E is made by OpenAI and is integrated into the chatbot’s interface. As a user, you tell ChatGPT what you need, and the AI assistant gives Dall-E the appropriate instructions. In the background, Dall-E creates different variants and presents one of them to you as a result.
You can ask for changes and improvements in the next step. However, ChatGPT and Dall-E will always create a completely new image. So you cannot ask to change just one detail.
Tips:
- Describe to ChatGPT in as much detail as possible what you need the image for and what purpose it is supposed to serve.
- Experiment with different styles, otherwise your images will look generic. For example, ask specifically for a “minimalist graphic” or a “3D render”.
Conclusion: Dall-E is perfect for beginners and users in a hurry, as there are no settings. The quality of the results is good. However, the lack of options and freedom is frustrating if you have a specific theme in mind. It works best if you let the AI inspire and guide you.
Stable Diffusion
Stable Diffusion is the name for a family of AI models from Stability AI. What sets them apart: They are open source and therefore freely available.
A very active community has developed around these image generators. As a result, there are several ways to use Stable Diffusion for free on your own computer – there are even apps for tablets and smartphones.
Stable Diffusion in its pure form is not beginner friendly, as there is a long list of options and features. Also, unlike Dall-E, you have to take care of the command prompt yourself.
Conversely, this means that once you are familiar with Stable Diffusion, you have enormous freedom. It is certainly the most powerful offering.
Tips:
- You can try Stable Diffusion for free here. If you want more options, you can find a demo of a more powerful web interface here.
- Also interesting is the Clipdrop collection of tools, which uses Stable Diffusion to remove the background of a photo, for example. Stable Diffusion Turbo, which generates images at lightning speed, is also very impressive.
- If you want to test it more thoroughly, use DreamStudio.
- For more advanced users, there are also browser-based user interfaces such as Automatic1111 and ComfyUI. For Apple devices, I recommend the free Draw Things app for Mac, iPad and iPhone.
- Find tips and tricks on the Stable Diffusion Art page (Note: not all images on this page are appropriate for offices and other public places).
Conclusion: Stable Diffusion has for me the feel of Photoshop and similarly powerful applications. As a result, it can be difficult to get good results in the beginning. But if you are not afraid of learning how to use it, you will get a very useful tool that you can even use for free.
Midjourney
Midjourney is popular, but I have to admit that I haven’t used it much. The reason: for a long time, it could only be used via the Discord chat platform, which I personally found off-putting. It made the process unnecessarily cumbersome.
In the meantime, Midjourney seems to be working on an easier to understand and more user-friendly interface. In this respect, I will certainly give it another try as soon as it goes live.
The results I see on the web can be quite impressive, but this also applies to Dall-E and Stable Diffusion.
Conclusion: If I want to generate an image quickly and easily, Dall-E is better and easier. If I want to experiment and have a lot of options, Stable Diffusion is the way to go. Midjourney seems to be very popular and seems to produce good results quickly. In this respect, this service may take its place between the other two in the future.
Further offers
The market is evolving rapidly. Here are a few more offerings in brief:
Meta/Facebook: Imagine with Meta is currently only available in the US. It generates four images for a text prompt and is free to use for the time being. Login is required. All images have a visible watermark identifying them as AI-generated.
Microsoft: Image Creator from Designer was formerly known as Bing Image Creator and uses Dall-E behind the scenes (see above). It’s a good way to try out Dall-E’s capabilities for free, if you can live with the sometimes rather strict limitations: The AI rejects even seemingly harmless image descriptions very quickly. It is integrated with Microsoft’s Copilot chatbot (formerly Bing Chat). It requires Microsoft’s Edge browser.
Google: ImageFX is currently in preview. Its special feature: It is designed to help users come up with image ideas and to experiment with different styles and motifs. That’s why parts of the text prompt automatically turn into pull-down menus. Use them to quickly select alternatives. All images contain an invisible watermark.
Amazon: The Titan Image Generator is currently available to users of the AWS Bedrock service.