The Evolution of a Portrait: Variation, Image Style Transfer & Photoshop
A couple of years before prompt-based AI generators were released to the public I was using Deep Dream Generator, specifically “Deep Style” to create images. This includes the one of Octavia Butler (above) that a lot of people love. Later, I failed in my attempt find the source of the original photo. A curator at the Smithsonian wanted to add it to a commission I was working on for the FUTURES exhibition in 2021. We had gotten image permissions for all of the other AI-generated portraits except this one.
Recently, I used the Midjourney Describe to remix the earlier image. This command that allows users to upload images and generates prompts based on the source images (see above). As a result, I had several thumbnails to choose from but each one had issues, especially with hands (as usual) and the facial features were not quite like Octavia’s. I found a few that had some promise, including this one I upscaled to get closer look at the details:
This upscaled image had promise but the fingers were impossible. The hair was too high and there were other elements that needed to be removed. This is where the Vary (Region) editor plays an important role. Vary (Region) allows users to select and regenerate specific parts of an upscaled image. This editor is closest I’ve seen to image remixing.
Knowing this image wasn’t quite what I wanted (still), I used the Pan option to lengthen the image, left Midjourney entirely and used Photoshop to make additional tweaks to the face and hair. I also edited parts of the background. Next, I used Deep Dream Generator to apply a certain style to the image. The style I went with is the same one I used in the original image (at the top). It came from a color-by-numbers app on my iPhone. Next, I composited both images in Photoshop.
Compositing is the process or technique of combining visual elements from separate sources into single images. A photo composite is multiple images combined or blended into a single image. Similar to collages and multimedia art pieces, composites can combine elements from a number of different pictures, paintings, graphic art, and other mediums. Image compositing is a skill that is usually separate from AI image generation, unless the plan is to extend the borders using Photoshop’s AI tools.
Although I used elements from the first (source) AI-generated and composited image, the final new version is unique. The final image is the results of several tools and multiple processes, as well as skills using image editing tools. Color knowledge is also important. It’s not about copying any one style, rather, the practice of remixing using GenAI aims to create something that no one tool (or human) can do alone.