Sophie Ashdown

I have an idea of an image in my head that I want to create, and I thought I could put a few different AI Image generators to the test to see what their interpretation was.  I selected three different sites and used the same prompt to generate the images:

“Black and white drawing of the London skyline in a constructivist style including varied patterns”

I chose to use Openart.ai, Craiyon,com, and Leonardo.ai.  The results are different of course but also surprisingly similar.

First I will show you the range of images that each site created, then select my favourite of each, then a final version that I will try to recreate myself…

Openart.ai gave me two results that are quite similar to each other but I like them both.  I like how the buildings are recognisable as London, there are a lot of patterns included, and it hasn’t gone over the top doing too much “AI weirdness.”

Looking closely at both images, the first one looks like a more representative version of London.  The second image has some AI weirdness that doesn’t really work.  For example there seem to be multiple domes of St Paul’s cathedral, and there are buildings in the background that I’ve never seen in London, particularly the transparent sail-like building on the right, and the beehive second from left.  The first image isn’t perfect either but with some adjustment I think it could work.  So this is my choice from Openart.ai:

Craiyon gave me nine different options.  These seem to be in a more simplistic style and whilst they are vaguely representative of London, some of the cityscapes look more like Copenhagen.  Perhaps if I just asked for a general cityscape then these images might have worked better, but apart from Big Ben popping up on a regular basis, these images are not obviously London.  Also the clouds are just terrible in all of them!

So I think there is not a “best” image to choose from the above options, but simply the “least worst”, therefore here is my choice:

Leonardo.ai is supposedly the most up to date AI tool of the three sites I chose, and you can see from the images that they do seem more advanced, almost more space-aged.  I feel like these images are the most obviously generated by AI, but at the same time have some of the most interesting combinations of the elements.  The buildings are more obviously London, although there are often multiples of the same building in one image.  I like the composition of the second image but there are three London Eyes.  I love the sky treatment in the third image, but it appears to have three Shard buildings. 

My choice is the fourth image as it has the most interesting composition and I love how it incorporates a boat on the river and hints of Tower Bridge.

So which is the one that I will attempt to create in real life?  It was mainly a choice between the OpenAI image and the Leonardo image but I have decided on this one from OpenAI:

As much as I like the Leonardo image, it seems too obviously an AI generated image.  The OpenAI option seems more in keeping with an aesthetic that I like and similar to a couple of things I’ve done before, so it’s not a million miles from what may have come from my brain.  I will need to make some adjustments and there will be a lot of artistic license along the way, but I’ll let you know how my version turns out…