Skip to main content
Thoughts from David Cornelius

Category

Roman ProgrammerThe fascinating realm of AI image generation has opened a new playground intersecting art, human creativity, and serious back-end computing. It's also opened questions of how we will tell real from fiction as this technology evolves and becomes even more sophisticated than it is already. For now, I'm just having some fun in my spare time!

From the vast store of uploaded images and video that grow by the billions every day, high-end computers using machine learning algorithms combined with large language models (LLMs), can turn a simple textual prompt into an image that is often quite realistic. For example, this image was generated with the prompt, "greek roman soldier wearing helmet and glasses sitting at a computer programming" and was generated using ArtSpace.ai. (I'm pretty sure he's a Delphi programmer, don't you?)

Guy in tee-shirt at deskBesides using prompts, AI image generation tools also let you base your new image on a theme or model. These next two images were generated using a similar prompt but different models; the one on the right uses one called, "Dream Studio" while the one on the left uses "Nova Flux 2024" (which is ArtSpace's newest and most realistic model).

Old and exasperated

Another factor at ArtSpace.ai, the tool I used for generating all the images on this page, is the seed. This can either be a number or (on some models), an image. I have not had success using an image to teach it "likeness" yet, I'm still learning the nuances of this technology. What is interesting is how very different images can be generated from the same prompt by simply using a different seed number. The following images were all generated using the same model and prompt but with one digit different in a 10-digit seed number:

 

Black & White Spiral with Color Tinges #1 Black & White Spiral with Color Tinges #2 Black & White Spiral with Color Tinges #3 Black & White Spiral with Color Tinges #4

Bikini girl on boat dock with third armAI image generators are not perfect. It can sometimes take several iterations of modifying your prompt to get the image you want; and it may actually lead you down a rabbit hole of exploration in a direction you did not expect. If you have a tool that lets you explore without time or cost limits, that can be a fun distraction. But it can be frustrating if you don't have the time to explore and don't know why the computer can't read your mind! I've seen a lot of example prompts that are long and explicit and generate amazing artwork. Learning to talk so the computer generates the image you want is a fine-tuned craft. It's all too easy to confuse the AI with text that seems innocent and clear to us but causes the generator to place an arm or shadow in an impossible position or not give enough fingers on a hand. I generated one of a woman in a bikini on a boat dock but in one of the iterations added the phrase, "motioning to join her in the water" and the AI added a third arm.

Large blue motorcycleAI image generators can also be used to envision products that don't yet exist or describe a home remodel taking shape or how new furniture would look in your living room. Colors and sizes and shapes can be changed and the image regenerated. I saw a video recently how retailers are using AI to help customers visualize themselves in new clothing or a different pair of glasses. To the left is an image created from the prompt, "large, futuristic dark blue motorcycle with chrome exhaust pipes, sitting in a showroom, oversize rear tire, blurry background". With a several more lines of descriptive text, an engineer could see potential problems with a design before spending hours in a CAD program.

You may have been playing around with some free tools or perhaps you've subscribed to an AI service to help in your programming and it happens to provide an image generator as well.  If you are still looking, I invite you to check out ArtSpace.ai which is one I discovered recently. It seems to be pretty good, has a lot of variety, the limit of 500 images a month is plenty for me, it's fairly quick (about 11 seconds per image), and I can download them from the web browser interface and use them any way I want--even commercially! Most software tools these days have a monthly or annual cost to use them and ArtSpace.ai has both; but they also have a seldom-seen one-time-price (which is just a little more than the annual price). For me, it was just too irresistible!

Computer in the shape of a sandcastle

 

david Sun, 09/08/2024 - 18:09

In reply to by Michael Riley (not verified)

You're welcome, glad you liked it!

I debated whether to write about this or not since it's not directly related to Delphi programming but I've been having fun and as soon as I started generating images, ideas came to mind about how these could be used to enhance blogs, to illustrate ideas, or come up with a cool, new avatar (like the one at the top of this page!).

But so far, it's just for fun; for example, I play board games with some friends and generated this one yesterday:

Four friends playing a board game

 

Add new comment

The content of this field is kept private and will not be shown publicly.
CAPTCHA
Please enter the characters shown; no spaces but it is case-sensitive.
Image CAPTCHA
Enter the characters shown in the image.