Variations on a theme.
After using Adobe’s AI Firefly to generate images in series from a single starting picture a few times, each time getting wildly different results, I thought I’d have a go at making a series on a theme. I started by trying out a colour theme, taking a small section of parrot from a previous experiment, and expanding it into a much larger square. One flaw in the system is that the new expanded image, while taking all the starting information from the centre section, doesn’t necessarily blend very well into what is already there. So I have to go back and repeat the generative fill over the original image to help it blend back into the new composition. Ideally, one day, Photoshop will have a kind of generative fill that takes stylistic information from a large area and alters specific content to match, because at the moment the only way to make the new image look cohesive is to generate over the original content, effectively erasing the shared origin. It’s kind of alright in the example here, being so abstract, but in a future post I’m going to talk about starting from an eye, and that is much harder to get right.
But still, I am in constant awe of the variety and complexity of the images Firefly offers. And I absolutely love how clues to where the generated content has come from present themselves. Paintings, sculptures, stained glass, painted silk, embroidery, play-doh, none of this was asked for, and I’m not sure that including such prompts would ever achieve the same result. I think finding that out will be a something I experiment with next.