Substack AI generated images
OK for illustrative purposes, seems ethically safe
When writing Substack blogs there’s a bit of a compulsion to include at least one image. Most of the homepage layouts work best when the entries they display include a featured image. But often my posts don’t really need an image. Or I haven’t got a copyright cleared image that fits. Substack deals with this by providing a good range of stock images - some cheesy, some OK. It also has an AI image generator. The results are not entirely bad, just about usable. The icon used for this site was generated using it - Shakespeare at a computer. I tried to generate “Nietzsche standing is space wearing a VR headset”, but it just gave me a generic image of a man with a moustache wearing the headset in space. Then I tried Socrates. Generic kind of ancient beardy guy. Jane Austen? It didn’t even recognise the name and just gave me a generic woman who might be called Jane. Same with Kate Bush. Finally, I tried “Victorian woman standing in space wearing a VR headset”, as you can see above.
It looks like there are good ethical boundaries being imposed. I couldn’t get it to represent real living people. Although it interprets them with stereotypical images, which may cause issues.
Here’s an AI generated image of “Dr Robert O’Toole of the University of Warwick dressed as a Cuban dancer.” The AI hallucination is much better than the real thing! But he only seems to be wearing half a jacket. Is that a Cuban thing?
