Thread by @KatGreenbrook

How I used #AIart to visualise text analysis data 🧵

This is something I’ve wanted to experiment with for ages! The data came from the London Stage Database as part of the @DataVizSociety's Data is Plural challenge.

Here's the final viz 👇

The data shows 140 years of performances between 1659 and 1800. I used text analysis to rank the frequency of words in performance titles - removing stop words (a, the, and, etc.). The top 50 words and their frequency formed the prompt word and word weight for the @midjourney AI.

@midjourney The @midjourney app runs on the @discord platform and uses complex neural networks to generate images.

midjourney.com/home/

I didn’t specify any design aesthetic because I wanted the data to “speak” for itself. Each query returns 4 image variations based on the prompt entered. You can choose to create more variations, or upscale them to add detail.

Some images the @midjourney AI generated were more abstract than others…

@midjourney I thought some images the @midjourney AI generated were incredibly beautiful…

@midjourney Despite the @midjourney AI generating the images, there is still a huge design curation challenge with this work. After a painful process of image selection, I used @Adobe Photoshop to edit and join the final images.

@midjourney @Adobe The top performance titles for the time periods analysed were added to each image in @Adobe Illustrator to provide context.

I use design elements to add emotion to my #dataviz all the time. But never has the data been able to produce these elements directly. I can’t wait to explore the #AIart field more. You can check out the final visual in more detail here: roguepenguin.co.nz/ai-generated-d…