Search All Articles

Browse by Category

Artificial intelligence is beginning to make significant strides. One of the most impressive is digital or AI art. I'll explain how you can use it or just play with this technology for free. The results can be amazing and useful. It might be something to add to your tool bag.

• Just as art!
• Create images for advertising, websites or newsletters. I'm using PlaygroundAI to create the images that liven up my newsletter, and I have negative artistic talent.
• As fun entertainment
• Illustrations for the graphically impaired
• Image beginnings for the graphically enabled
• Art for the graphically impaired

You give it a written description of what you want. Then the AI reads your description and tries to fulfill your prompt with an illustration. You can then take that and alter your description and continue to refine your image.

Here's an image I created to illustrate the idea of computer generated or digital art.

I told it: photorealistic image of French robot who is painting a picture with canvas and easel, photorealism, intricately detailed, 8k

I'd imagined the robot wearing a beret, which perhaps I should have specified. The computer seems to have missed my whole French idea. But I really liked the 3 dimensional aspect to the painting it was making. This took many steps as it got closer and closer to what I wanted and I tweaked my prompt as well. This was done on Playgroundai.com for free. I could have added "wearing a beret" after a robot if I still wanted that after seeing this.

This was what got me excited about digital art. I saw this image from John Scalzi, one of my favorite Sci-Fi authors. He posted it to Mastodon.

He told it: The Snowman Regrets Having to Melt

He also posted this:

He used the prompt: In His Cage the Lion Dreams of the Sky, in the style of Van Gogh

I believe John Scalzi uses https://midjourney.com

You can see a showcase of digital AI art from midjourney.com here if I've managed to whet your appetite. You can see the prompt used by pointing to the image, but you can't enlarge it and get more information without an account. Playgroundai's showcase of recent images is here. With theirs, you can not only see the prompt by pointing to it, but click on it to increase the image size and see more information. This helps you figure out how to talk to the computer's AI.

These tend to be slanted toward Sci-Fi and Fantasy, which, I suspect, is the kind of art many people drawn to this technology like. It is not restricted to that. As the tech spreads, we're getting more diverse pictures.

I created this image to illustrate the article on blocking unwanted popup messages later in this newsletter.

I did it with the free playgroundai.com site using stable diffusion 1.5 and this prompt: Man at computer, frustrated, upset, needless interruptions, Cartoon, Indoor, office setting, very frustrated, style Shinji Aramaki style Inio Asano, detailed.

I wanted to do this in an anime style and found two artists who do work in that style and told it to use their style. Perhaps I could have just said, Style anime. Again there were many iterations with the AI as I tried to get the character sufficiently frustrated.

Last month I did this illustration for my article on how to listen to podcasts on your phone.

I prompted it with: Happy Panda, with headphones, in garden and chose a floral background. I used the very simple and limited https://dream.ai/create. This was before I discovered Playgroundai.com.

There are three popular models, Dall-E, Stable Diffusion and Midjourney. Dall-E is from Openai.com and it is an amazing company. However, its model is proprietary and closed. It also tends to be somewhat more expensive to use. Stable Diffusion on the other hand, is open source and potentially free. All three models are amazing. I think non-tech people will find Midjourney too hard to setup. It can also get expensive pretty quickly. So far, PlaygroundAi gives you the most for free and the clearest paid version pricing. It seems the least expensive as well. It offers free use of Stable Diffusion and $10/month for 800 images with Dall-E 2. Paid accounts of Stable Diffusion are $15/month and offer bigger sizes, and more detailed designs.

One part of the process is for the AI to parse what you type and "figure it out". Improving its ability to "understand" our prompts is an important area for training. Then, trainers go through millions of pictures and labels to teach the AI what labels go with what pictures. The AI learns to alter images and mesh them with what you typed. These models require tens of thousands of hours of training on what sort of descriptors go with what pictures.

I think Dall-E does a better job than Stable Diffusion in "understanding" your prompt without constantly changing the prompt and working to get it right. They tell me that Stable Diffusion does a better job of letting you tweak your images afterwards. I'm not good enough to speak on that.

In either case, you go to a website that uses one of these models and has trained it. Then enter your prompts. It'll generate some images, then you tweak your prompts and other aspects of the picture. As you go through stages, you tell it to work from your best picture to date and make some changes.

Processing your prompt and generating detailed images, then repeating the process as you tweak your prompt, requires a lot of processing power. Furthermore, we want it to be fast so we can choose an image and then tweak our prompt and generate a few more options. We want our art stored, so we can go back to it. All this costs the website money and they need a financial plan, even if they use the open source stable diffusion model. This technology is so new, these models are changing because no one knows what will work and what won't. Something could be free one month, and cost the next. Everyone is just trying to figure it out, both us and the websites.

I chose Playgroundai.com. It'll let you do a lot of art for free, so you can figure things out without spending a penny. Everything you generate you own. With some sites, you have to give the site credit for everything you generate unless you buy an extra expensive account. PlaygroundAI gives you ownership even with the free account. Other sites make you pay if you use the art commercially, and that includes a simple blog, or this newsletter, although it is free.

After using PlaygroundAI I was so impressed that I wanted to support them and bought an account although my limited use didn't seem to require one. They'll probably be forced to tighten some free parameters, but their heart is in the right place and they are doing a great job.

Top of left side

They have an area for you to write your prompt. Then any objects you want removed from the generated art on the next round. Perhaps it put a moon in the background you didn't like, or made it too red.

The filter gives examples of different styles you can use, and they'll generate the prompt for you to create that style. Here are some preset styles you can use with Stable Diffusion 1.5

You just select one, and it'll create the appropriate prompts. In the middle of the window is your creation. To the right are other options.

Here you choose the size of your image, how strictly the AI is to follow your prompt versus the freedom it has to create. Quality and details are how many times it'll go through the image and work to improve it and create finer details.

You can see why I appreciate these prompts available for me, without needing to figure out the prompts for each one. This makes PlaygroundAI easier to use than some other options.

• Playgroundai FAQ
• PlaygroundAI 15-minute YouTube Tutorial:

Date: December 2022