As a graphic artist, it is important for me to stay on top of the many tools that are available to create and deliver thoughtful design. Since the options for generating digital design have exploded in recent years with the advent of generative artificial intelligence (AI) and nonfungible tokens (NFTs), the task of keeping up with everything can feel daunting.
My personal method of combating the feeling of being overwhelmed is to do research to try to understand how things work relative to how I work.
I thought I would share a practical example of how I used two different publicly available generative AI platforms, WOMBO’s Dream and Fotor’s GoArt to enhance one of my own design experiments. I will share my experience with NFTs at a future time.
The starting design
For my generative AI collaboration, I started with an original photo collage, which I created by combing three photos together using Adobe Photoshop.
I do a lot of design practice by creating editorial compositions from my own prompts. In this case the prompt I used was “When you need to work, but your mind is somewhere else.” I feel this way often.
Here is the starting photo collage:
The initial result is a bit “out there” but it captures the emotion I have about the text. I like plants (my thistle photo), live near the beach (my beach photo) and my head can be in another universe (cosmos stock photo). When I am not working, I like to be out in nature.
Not everyone may feel the same way about this prompt. A good way to guage if my interpretation of this sentiment resonates is to post the collage on social media.
The problem with social media is that feedback can be limited by the size of one’s network, interest of an audience and the algorithm of the social platform.
For more directed conceptal feedback, I decided to use WOMBO’s Dream generative AI platform to see what it would create from my prompt.
Step two: Concept testing with Dream
The Dream generative AI platform uses a combination of CLIP, a neural network that connects images and text, and VQ-GAN, a generative adversarial network (GAN) that doesn’t contain inductive bias (check out Alberto Romero article for his take on the potental of these types of platforms to redefine art.)
The Dream web app allows you to insert a text prompt of 100 words or less and infinitely generate images from various styles to create original art (and soon, mint outputs as NFTs). In my case, I wanted to see how Dream would interpret my “mind is somewhere else prompt”
I used my prompt to cycle through the 20 styles offered and chose the following three as my favorites:
I decided that the outputs were pretty cool and interesting in various conceptual ways, but none of the compositions fit my personal style or the emotion I wanted to convey. Everything generated looked heavy not daydreamy.
Most of the images do include visual impressions of common work-associated items like clocks, desks, post-its and notebooks that are missing from my photo collage. If I were creating something for a client, this type of conceptual information is very insightful.
I still wanted to improve my original collage with more deepth of styling. Next, I decided to try Fotor’s GoArt platform.
Step three: Stylizing with GoArt
GoArt is a generative AI platform based on neural style transfer, which imparts the style from one image to the content of another image. There are many neural network models that perform transfer of art style (see this review article I found). The GoArt interface is easy to use and offers over 50 art styles to chose from.
I uploaded my photo collage and cycled through every available art style. Here are three I liked most for stylizing my collage:
The French Coffee style transfer effect fit the best with my collage’s original color palate; however, the styling change made the image feel wintery.
I liked the texture the Colorful World style transfer effect imparted to the design, but the color palate and contrast were too dark.
I decided to move forward and use the Rich Colored style transfer effect to create my final image because the bold oranges and blues fit with preffered palette and previous creative works.
GoArt lets you download a lower quality, watermarked image for free or sign up for a free trial to get premium image options. As you can see by their watermark, GoArt has positioned themselves as a NFT creator.
Final Step: Personalizing the Rich Colored GoArt image
To create the cover image of this article, I used the Rich Color GoART image as input in both Adobe Illustrator and Adobe Photoshop to vectorize, smooth, color retouch and reformat for the final product.
First, I think it is great that Dream and GoArt are positioned to help people to create original art without the need to know too much about art. The freedom to create should be more accessible.
I shared this example to provide some insight on how to use generative AI as a collaborator to improve/vary original art work. As a freelance artist, it can be challenging to find a collaborators and artifical intelligence helps fill a gap.
Thank you for any feedback and please share ways you may have used generative network tools to enhance your original art.
Julia Fletcher is founder of JEFS Storytelling Arts, a graphic design studio, where she uses her unique research skills and artistic talents to create custom visual stories that help clients’ increase engagement and promote the education of their audience.