I got my invitation from OpenAI DALL-E a few days ago and decided to play with it a little bit.
What is OpenAI DALL-E?
DALL-E is an AI system that can generate realistic images from a description. It uses a process called diffusion to combine different objects in the same image in a cohesive way. It started as a research project and is now available in beta. You can join the waitlist to get an invite.
Using the OpenAI Front-End
You are met with a simple user interface when you sign up.
You can now get creative and generate whatever you would like to see. In the following example, I entered a puppy and a kitten playing in a green field and pressed the Generate button.
Here’s what it came up with:
I think the results are quite impressive. If I didn’t know they are AI-generated, I wouldn’t be able to tell so. There are some small giveaways, but I wouldn’t read too much into it.
If you hover over the images, a button appears which shows more options:
You can create even more similar images from it by clicking Generate variations. For example, the variations of the above image look like this:
Another option to generate images is uploading your own image.
To test this feature, I uploaded a stock image:
Accepted the disclaimer:
Then I was asked to crop the image.
I accepted the default suggestion and clicked Done.
Then I asked to pick an action:
If you click Edit image, you are taken to an image editor. Looks like all it does is remove some parts if you like, though.
I then went ahead with Generate variations action. Here are the results:
It doesn’t look like there are many variations, but then again, maybe it’s too simple to allow variations. I guess at this point, you just have to play with various ideas and see what works best.
Using DALL-E Programmatically
I wanted to write some code to generate the images programmatically, but I couldn’t find anything about it in the OpenAI docs. I asked the question on their Discord server, and a bot answered my question:
So looks like that’s not an option for the time being.
Conclusion
This is just a quick overview of the OpenAI DALL-E project. There are other projects out there that produce similar results. It’s a very active field these days, so it’s worth watching the space. I’ll likely post an update when the API is available.
