Last week I talked about my experiences using a variety of different AI art generation tools, and this week I've finally got access to OpenAI's incredible new Dall-E 2 system.
Sign up now
Before you go any further, sign up to the waiting list for access to this system right now, because I guarantee that by the end of this blog you're going to really want to play with it. I had to wait several months to get my hands on it.
Dall-E 2 is the latest in a series of tools that take text prompts as their input and create output images. I've been seriously impressed with some of the systems, but they usually ran pretty slowly and included strange artefacts that I had to remove in Photoshop before I wanted to show them to anyone. But not anymore. Dall-E 2 absolutely knocks it out of the park - the images it generates are (for the most part) indistinguishable from human-created art.
Spot the AI
Here's a quick game: can you identify which of the pictures below were created with Dall-E 2 and which are random pictures I stole from the internet?
Obviously they're all Dall-E 2 images, I wouldn't steal random pictures from the internet 😱
So what's it for?
During the sign-up process, I had to agree that I wouldn't use these images for commercial purposes, so really what's the use of this tool?
Having played with it for a couple of days, I've found it most useful for generating ideas - I'm currently creating a digital artwork based around a mountain scene and I've been using Dall-E 2 to generate scenes that I can use as inspiration for the artwork - taking a colour palette from one image, shapes of the mountains from another, a detail from a third. Here is a selection of the images that it created:
We're currently designing a logo for a new client's counselling business and we've been using Dall-E 2 to come up with quick concepts for the logo. The designs it creates aren't tidy enough to use commercially even if we wanted to, but they give us a great place to start a conversation with the client about what they like and don't like:
The final thing I wanted to share today is the ability to upload an image and have Dall-E 2 create variations on it. Let's say you're doing some marketing for a cat food company, but the images they've given you don't quite fit in the space on their marketing material? No problem, upload a picture and Dall-E 2 will create variations on it, like so:
This raises some interesting questions about the use of image libraries in future, where people use copyright images as the starting point for AI-generated images that don't have copyright restrictions.
Don't get me started on using this to create deep fakes - luckily the tool doesn't currently let you upload images that include faces.
I hope this has inspired you to sign up to use Dall-E 2 and start playing with the system. It's easy to use, generates coherent, beautiful images quickly and the limit is really just your imagination.