Using AI to generate images for blog posts

I've been doing a lot of reading around Artificial Intelligence recently, it's a subject which fascinates me. As well as the ethics of AI, as both an online content creator and consumer I'm concerned with how AI could be used to produce content which is indistinguishable from that produced by humans. I've played around with ChatGPT to both generate blog post titles and to see what posts it would write based on those titles, and I've been impressed and I must admit a little bit in awe.  

When I'm looking for an image to illustrate a blog post then I turn to Unsplash for good quality and copyright free photographs that work with my content. The other day I was struggling to find exactly what I wanted, and I remembered reading about the AI image generator Stable Diffusion. I gave it a go and was actually really impressed with the image of a packed suitcase that it generated (the post is to be published next week, the AI image will be clearly disclosed!) so I thought I'd have a go at generating some 'stock' photographs that I could use to illustrate the sorts of things that I often write about!

I started with some crafty requests - bullet journals, art journals and a crochet blanket - and I was pleasantly surprised with the results. The text in the journals is nonsense and the blanket is a little fuzzy, but as part of a quickly skimmed blog post I think that they could pass as real photographs.

I soon discovered that anything involving text or body parts like faces or hands doesn't work well at all, and I ended up with some very creepy images. I was impressed that the tool managed the word 'blog' on the laptop, but all the hands have gone very wrong!

I found that the AI image generator coped much better with images that involved a landscape or food. The images are a little too perfect, very smooth and shiny, but I think that some of them could certainly pass as genuine photos when scrolling through a platform like Instagram!

My favourite AI generated images were these two involving books. It took a few attempts because the tool can't manage text very well so I needed to try for images taken from a distance or a little out of focus. I think that either of these two images would make a great blog post photograph, as long as you don't look too closely!

Finally one of my favourite images was one that went completely wrong. I asked for a person sitting on a sun lounger at the beach reading a book and ended up with this disturbing image!

Clearly the technology has a way to go, but I can see that it won't be long before it will be easy to generate the exact image that you are looking for to illustrate something online, especially if you are just looking for generic images. I'll be looking very carefully at photographs now, because I've started to identify the characteristics that you can look for to identify an AI image. For example differences in the textures, wobbly edges, wonky text, over saturated colours and sometimes a rather grainy feel.

I found it fascinating to play with this tool, and if you have a spare few minutes I'd recommend giving it a go, you might be surprised with how well it works! 

All these images were generated using Stable Diffusion which is currently free to use. All images created using Stable Diffusion Online are fully open source, explicitly falling under the CC0 1.0 Universal Public Domain Dedication.