Adventures with Artificial Intelligence images

I have had considerable success with Artificial Intelligence tools. Most of my success has been inside Adobe Bridge and Photoshop, but I also have had tremendous success with Chat GPT for publishing a catalog from an out-of-print booklet. Click here to read how that project was done.

I’m most impressed with Photoshop’s AI Noise Reduction. That one is the most productive for my photo work, and it is the one that has me gushing in praise of what AI is capable of. You can read my praise of AI Noise Reduction here.

And, I have also been toying with some of the AI engines to create images from digital whole cloth (whole pixels?).

A few months ago I purchased the blognosticator.com domain, after years of trying, and I wrote about that in a previous blog post. On that post I needed an illustration of a man putting up the .com suffix. I started by looking at stock images of men putting up things, then I used one of those images as a model to draw my own man putting up the .com suffix. I made it look like a blueprint (a technology largely unknown to most people these days). But I liked the motif so I did it that way. Click here to see that illustration as I published it.

After completing that post, I moved on, working on my building permit application, getting ready to teach again (I’ve been hired back at Cal Poly for the year), and going off to Burning Man again (no rain this year).

But there was a nagging thought in my head… could I have asked an AI engine to draw the man putting up the .com suffix? I decided to give it a try.

Attempt No.1: Chat GPT: I put in a prompt asking for a man putting up a poster, the man’s back to the viewer.

Response: Chat GPT told me it can’t create images.

I tried Chat GPT 4.0 today, and got the same response.

Attempt No. 2: Google Gemini: I entered a prompt asking it to:

Create a photo of a man putting up a poster on a wall. His back is to the viewer, and he is wearing a hard hat.

I received this response: Generating images of people is only available in early access with Gemini Advanced. Get early access to new Gemini features when you subscribe to Advanced here. I signed up for the free trial period and proceeded.

I tried, and succeeded in getting a photo quality image of a man putting up a poster. With “refine” prompts I was able to get Gemini to include more of the ladder. In the end, I was pretty happy with those images. Here is my favorite:

I liked this one the best. It allows me to have the man installing the .com suffix on a sign.

When I tried, as a stand-alone assignment, to get Gemini to draw the letters .com as a large sign on a building, I didn’t fare so well. I got Gemini to render the word “blognosticator” in lower-case letters as a sign on a building. And, when I asked for the .com, it succeeded. I tried several “refine” prompts, each one got worse. Eventually I stuck with the first one, which is OK.

I put the man putting up the poster into that photo. He’s putting up the .com suffix now. It turned out pretty well.

I ended up not using the ladder at all! I put the man into the AI-generated sign of blognosticator.com. Both contributing images were made by Google Gemini.

My love of street art mixed with Artificial Intelligence

I have recently attempted to get each of the AI engines to draw the word “BLOGNOSTICATOR” on a red brick wall in graffiti style. I figured that this would be pretty easy for these brainiac applications. I was very wrong. It turns out that Chat GPT does words really, really well but not photos. The other AI engines also handle text quite nicely. But the photo-generating part of each one has a severe problem with spelling. I would have thunk that spelling would be easy for these programs because letters can be put into a matrix and then rendered.

I started in Adobe Photoshop, using its Generative AI tool. I put in the following prompt:

Please make a photo of a red brick wall. On that wall is the word BLOGNOSTICATOR in graffiti style, with bright colors and clever lettering.

Photoshop chewed on that for a couple of minutes and then presented this:

Badly spelled, but OK otherwise. I wasn’t thrilled.

Curiously, the word “BLOGNOSTICATOR” was misspelled. I thought it might be a one-off error, and added double-quotes to the word to get the AI engine to understand that it was literal. Usually problems in text processing or computer-learning are improved by specifying words that are not in any dictionary as literals.

And, Photoshop’s AI failed again. And again. And again.

I tried simpler words to see if it would misspell every word. I asked it to paint the word “STITCH” on the wall, and it came back spelled “STCH.”

Really?

I switched to Microsoft’s Copilot software to see if it could do the BLOGNOSTICATOR image, or any image with words.

Copilot fell on its digital face with BLOGNOSTICATOR, but it did succeed with the word APPLE. I was thrilled, but I didn’t need APPLE. I asked that application to generate more images of the word BLOGNOSTICATOR on a red brick wall.

Of all the AI systems I tested, Microsoft’s Copilot produced the most pleasing graffiti style illustrations for me – despite the misspellings. The three images it created are stunning. I love them. I saved them and opened them in Adobe Photoshop and edited the word BLOGNOSTICATOR by hand so that it is spelled correctly. Now I love these images even more. They are creative, showing real style, where the other engines I have tried produced more pedestrian images. They produced good images all the way around, but none showed the level of “creativity” that Copilot did.

If only my blog were called BLOGNOTISTTOR! I love the art, but the spelling is terrible. This, and several other extraordinary images were generated with Microsoft Copilot.
With some hand modification in Adobe Photoshop, I was able to make this image perfect, and I LOVE the artwork! This exceeded my expectations.

One image produced by the Adobe Photoshop AI system proved that Adobe’s software is also capable of “creativity.” I asked for the man putting up a poster. The first result is incredible!

The man is beautifully rendered, and the composition is excellent. What showed “creativity” is that the man has drawn a self-portrait on the poster (I didn’t ask for anything to be on the poster). It’s by far the best AI illustration that I have received yet and the one that shows computer introspection.

This is my favorite image so far. It was created with Adobe Photoshop’s AI Generative Fill tool. I have not modified this image at all, except to reduce its size for this blog.

As I learn more about these tools, I will post more. Please stop by from time to time and see what I have wrought – by typing prompts on my keyboard. And, if you would like to be notified when I post new stories, there is a button way down at the bottom of this page that you can click to be informed of new posts. I have over 450,000 readers now.

About Brian Lawler

Brian Lawler is an Emeritus Professor of Graphic Communication at California Polytechnic State University, San Luis Obispo and was a Guest Professor at Hochschule München from September, 2021 to September, 2022. He writes about graphic arts processes and technologies for various industry publications, and on his blog, The Blognosticator.
This entry was posted in Adventures, Art, Artificial Intelligence, Technology, Typography and tagged , , , , , , , , . Bookmark the permalink.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.