DALL-E 3 vs OpenDALL-E

Hey, what’s up?

Today, I wanted to test and compare DALL-E 3 and OpenDALL-E v1.1 to see how they currently differ.

To do that, I’m going to perform 3 tests:

Photo realistic image
Vector Image
Text Image

If you’re looking for an alternative to DALL-E or if you’re considering free and open source AI models, then this could be for you!

Let’s go right in!

WHAT IS DALL-E 3?

Dall-E 3 is a popular AI image generator developed by OpenAI. You can currently use DALL-E 3 from many places like ChatGPT, Microsoft CoPilot or Microsoft Designer. Depending on the service that you’re using, you might have to pay a monthly subscription and/or give away your privacy. That’s usually what happens when you use such online services.

For these tests, I will be using Microsoft Designer which uses DALL-E 3 underneath, because it is currently free to use.

WHAT IS OPENDALL-E?

OpenDALL-E v1.1 is the AI image model developed by the open source community, more specifically DataVoid from CivitAI. It is based on SDXL 1.0 and its purpose is to as close as possible to DALL-E 3. Which is why I wanted to compare both because I’ve been using DALL-E 3 a lot in the past.

To use OpenDALL-E, you need to download the AI models and use a software like Automatic 1111 Stable Diffusion to generate AI images using your GPU. This means that you need a decent computer. In my case, I’m a PC gamer and I love having a powerful computer, so that’s not an issue.

SETUP & SETTINGS

Before we start, I simply wanted to configure my setup and settings properly.

On the left side, I’ll be using Microsoft Designer.

On the right side, I’ll be using Automatic 1111. In this case, I am a few more settings. I’ll be using the settings recommended by the creator, DataVoid to generate these images.

CFG Scale: Use a CFG scale of 8 to 7
Steps: 60 to 70 steps for more detail, 35 steps for faster results.
Sampler: DPM2
Scheduler: Normal or Karras

1- PHOTO REALISTIC

The first test that I wanted to try is a photo realistic image to see the overall quality of the images generated.
To do so, I’ll be using the prompt:

cool monkey with sunglasses surfing on a big wave in Hawaii, photo realistic

I think that the styles are quite different. They are both generating awesome images and prefer a bit more DALL-E 3.
The style of OpenDALL-E seems a little more cartoonish and has a plastic feel to them.
DALL-E 3 seems to have much more details.

2- ILLUSTRATION VECTOR

The second test that I wanted to try is a vector image because it is usually better to work with these kinds of images for designing t-shirt and print-on-demand designs.
I usually prefer specifying vector images for my print-on-demand because it is usually easier to use a tool like an autotrace to edit and make changes to the images. That being said, I hope one day that I’ll be able to directly generate SVG instead of PNG or JPG.

To do so, I’ll be using the prompt:

illustration of a cute orange kitten with big green eyes, vector image, white background

Again, I think that the styles are quite different. They are both generating awesome images and prefer DALL-E 3.
The style is of DALL-E 3 looks more like an illustration for children. They are also much more detailed.
OpenDALL-E managed to generate illustrations, but it is still quite photo-realistic.

3- TEXT

The third test that I wanted to try is a text image because text based designs are also pretty common when designing for t-shirt. I need to make sure that the AI is able to understand and generate the text properly.

To do so, I’ll be using the prompt:
“keep calm and reboot for the AI revolution” with a crown on top, black text, white background

I think that they are both able to generate text, but tall have mistakes. That’s where we are with AI.
But overall, DALL-E 3 seems to be much more detailed and clean than OpenDALL-E.
Open-DALL-E is more simplistic.

CONCLUSION

In conclusion, I think are quite many differences between the real DALL-E 3 and the free and open source alternative OpenDALL-E.

Overall, DALL-E 3 seems more versatile in the way that you can generate many types of illustrations, better photo realistic images and more generally more detailed images.

That being said, you might have to pay a monthly subscription for DALLE-3. For now, Microsoft Designer is free to use, so might take advantage of before they lock it behind a subscription.

You can use OpenDALL-E for free as well. But you need a decent computer, the proper installation to do so, but you’re able to generate everything yourself locally with full control. It also takes a little more time to generate. In my case, that’s not an issue because I’m a PC gamer and studied computer science. Having a powerful computer is an important thing for me.

Another thing to keep in mind is that DALL-E 3 will surely receive updates in the future because OpenAI is actively and working on it.

With OpenDALL-E, it actually depends on the creator of the AI models because he is probably doing this project in his free time. Usually, future updates are never assured from the open source community, but there are many other AI models to choose from.

If you’re going to use OpenDALL-E, make sure to give a like to the creator or contribute in any way because I think this is great work.

So that’s pretty much it, I hope this was helpful to you. If this blog was valuable to you, you could consider following me on social media and if you have any thoughts or comments, you can leave them below.

I hope this helps, peace!