I am editing an anthology of AI co-created fiction. This is due next month. During this I experimented with text to image using NightCafe and now Midjourney. I won’t go into the technicalities (use search) but prompt programming is the key.
It is a lot of fun and is a form of ‘casual’ or ambient creativity, but can also be used for mock ups and start work etc, or to get ideas for further human artwork.
The system makes 4 images as below, and then you select and upscale the subjectively best one. This is how the ‘AI’ is being trained or improved, via human quality selections.
Midjourney system – prompt was “Computer art in the style of Micro Arts Group Geoff Davis”
Midjourney system – prompt was “Computer art in the style of Micro Arts Group MA4 Story Generator” – this was a text generator, so it gets the monochrome right, and one has a keyboard.
Midjourney system – prompt was “Computer art in the style of Micro Arts Group Martin Rootes” – this has the BBC Micro in it too, must have found that in the training data