My Top 10 AI Images of 2023 - - - - - - - - - - - #4 The Hanged Man challenge

AI excels in numerous areas, but it also has its limitations. These limitations can be quite frustrating, particularly when you consider how AI can effortlessly generate incredibly detailed images within seconds. Yet, we still encounter peculiar errors and anomalies. As of this writing, some of the most common challenges include: generating coherent text messages, depicting realistic human hands, simulating archery skills, and my personal favorite, recreating "The Hanged Man" card from a traditional tarot card set.

This might be my favorite image of my Top 10. Quite a hauntingly beautiful image.
I added details of him being infected by a fungus - more on that later . . .

I stumbled upon this AI limitation in a Reddit group chat. A user was expressing their frustration with attempting to create a tarot card set using AI-generated images but faced a significant obstacle – they couldn't get AI to depict "The Hanged Man." Many of us in the group, including myself, attempted to generate the image, but no one came close to the traditional depiction of the card. (Which typically features a man hanging upside-down by a rope tied to one of his legs, the other leg bent horizontally, with the other end of the rope secured to a tree branch above.)

An example of the classic Hanged Man tarot card

Some argue this happens because the AI is censored from showing a hanging person. Others said, AI doesn't quite understand what "upside-down" means and still needs to learn about terms of placement, locations, directions, and alignment. I agree with the latter. Often you can not direct AI where you want things to be placed in a text prompt, it's very hit-or-miss.

My first attempt at The Hanged Man (left) and the flipped version (right)
The gravity-defying skull chains, lowered branches and bottom light source ruin the hanging aspect.
Even though he has two left hands, I liked this image enough to use as a reference for the top image.

Certain individuals believed that a potential solution to the problem was to create an image of a hanging man and then flip it upside-down to achieve the desired effect. However, this approach proved to be ineffective. This is primarily due to the presence of elements like the ground, trees, and tree branches in the background of this particular tarot card. Additionally, AI algorithms incorporate factors such as gravity's impact on clothing and the positioning of upper lighting sources in their image generation process. Consequently, when you flip an AI-generated image, it often still appears as if you simply flipped the original AI image. 

My Top 10 AI Images of 2023 - - - - - - - - - - - #3 Red (Not So Little & Still with a Hood)

For years I have been developing a character for a children's book, directly linked to Little Red Riding Hood, a character deeply embedded in fairy tale lore and possibly the most iconic in its history. This story, familiar to many, has undergone numerous revisions and transformations through the ages.

Reimagining fairy tale characters isn't a novel concept. However, I often ponder about the lives of these famous characters beyond the "happily ever after." What happens after their story?

The Lovecraftian arcane world of a more mature Red Riding Hood

I envisioned the Red Hood child evolving from her iconic tale to become adept at protecting herself against any creatures that might speak, threaten, or disguise themselves as innocent grandmothers. In a Lovecraftian world filled with various mystical threats to humanity, I could see "Red" mastering the mystic arts, equipping herself to face and overcome these challenges.

The image above was generated only due to the new "Outpainting" feature in MidJourney at the time. "Outpainting" allows the user to expand outside or beyond the edges of an image and build up an image with greater details and background. 

This was the original Leonardo.ai image before using the outpainting feature to expand the image 

Before doing the top image, I wanted to experiment with training a "model" in Leonardo.ai for the AI to better understand what my "Little" Red Riding Hood" character should look like. NOTE: A "model" can be anything: an animal, wooden chair, forest, tall building, etc. . . (not just fashion styling pretty ladies.) 

My modernized version of the famed Little Red Riding Hood
This is one of the images I created to represent the look of the character

To train a "model", one collects and uploads images that closely resemble the subject, ensuring the photos are of similar size. After completing the training phase, this model can be loaded into the AI system, which will then have a general understanding of the character's appearance.

Little Red Riding Hood, no longer little and now masters the mystic arts.
In my story draft, the iconic "Wolf" was in fact the leader/father of a pack of demons.

My Top 10 AI Images of 2023 - - - - - - - - - - - #2 Willie Wonka Kaboom

I've been working on this kid's book draft for years now; directly inspired by Roald Dahl's "Charlie and the Chocolate Factory", and the original classic movie with Gene Wilder as the titular character. I tried making some Wonka images. The final result was way cooler than I thought it'd be. 

Yes, the candy parts are a bit vague, but the whole 'candy explosion' was great. I do wish I could've thrown in some floating Wonka bars, but AI's still iffy with logos and fonts and placement of items.

Something has gone terribly wrong with Wonka’s candy factory.

At first, the images generated weren’t looking like the Gene Wilder Wonka I wanted. Many weren’t truly Wilder-like at all. These images were mimicking Wilder’s features, with nothing relative to the character. Anyone who uses AI knows, it takes time, patience, and a bit o’ luck to get the images you hope to generate. So I had to learn how to keep building up details, eliminating what wasn’t working and adding to the description what was missing.

Here are a few of the original terrible results . . . 

1st attempt was so bad and dreary - especially in grayscale. (What’s the blue container in front for???)

This one was so manic; Wonka looks as though he could kill, (or maybe he already hid the body?)

Again with the grayscale, feeling like a depression era image, with a heavier Gene Wilder

My first AI Image (Precursor to My Top 10 AI 2023 list)

It's been less than a year since I began exploring AI-generated imagery. My journey started when I joined MidJourney on February 16. For my initial attempt, I asked my wife for an actor and character combo; she suggested Ryan Gosling as Mr. Rochester from "Jane Eyre." The result was my first MidJourney creation.

My first generated image with MidJourney

A half-minute later, several images following the prompt description appeared. Everything I knew about creating and sourcing inspiration was about to change. In the following weeks, Midjourney would upgrade to version 5.0, drastically altering the level of image detail. I absorbed as much information about AI I could find, and started seeing ways to use it as a major tool for my work as a Product Designer. 

Eventually, I added Leonardo.ai and ChatGPT to my roster a month later, along with many other free AI services. and tools. Improvements, methods and tools were popping up every week. 

But enough history, let’s begin the 2023 list . . . head to the next post!

Santa Design 2023 (AKA Is Adobe Fresco good?)

Merry Christmas! . . . and a Happy New Year!

Santa character for 2023

Every year, I draw a Santa (or a few). This is my personal annual challenge: to draw Santa differently each year, never repeating the drawings from any previous year.

This year, I created a Santa for the first time using Fresco, Adobe's competitor to Procreate on the iPad. Like Procreate, Fresco boasts an ideal interface and excellent control over drawing lines with a stylus like Apple's Magic Pencil, or on a Wacom screen. Note: Unlike Procreate, which is currently only available on iPad, most Adobe products, including Fresco, are available on all platforms, including Windows.

I would almost say my line control feels even better with Fresco. Depending on the smoothness settings, Procreate can sometimes alter a line's path, especially near the end of a brush stroke when lifting off the tablet.

The best thing about Fresco, to me, is that you can paint and draw in Vector mode. This means you can freely resize your drawing to any size without losing detail or bitmap blocking; all lines stay smooth. It's well worth a look, especially for any t-shirt designers or poster printers out there. I may find myself weaning off Procreate due to this excellent program... (Check here next year to see which program I use for 2024's Santa!)

Dynamic and Explosive AI Character Developing

- - - - UPDATE - - - - (Posted on: Tuesday, October 25, 2023)

I meant to show the explosive level of creating a character with powers.
I truly lucked out on this image; so powerful and filled with emotion. I do feel like this image used some reference to Jean Grey of the X-men, the outfit and long red hair seem to echo that character throughout the image. For those interested, here is the prompt and settings use:

Using Leonardo.ai • Dimensions:744px x 1320px • Resonance:15 • Contrast Boost:0.5 • PhotoReal:On • Alchemy:On • Preset:Cinematic • Raw:On • Field of Depth:Low • Seed:963356160

Prompt: A Photograph capturing a woman's explosive fury in a comic book style. Show her intense anger radiating from her body as energy bolts crackle around her, her eyes ablaze with fiery rage. Use dynamic lines, bold colors, and exaggerated facial expressions to convey the intensity of her wrath.

- - - - UPDATE - - - - (Posted on: Tuesday, October 24, 2023)
Finally, after several attempts, I was able to generate the image I wanted. The image depicts a female character with deadly psionic daggers, glowing green eyes, and short red hair. She wields her daggers with precision and skill. Creating an image can be a time-consuming and challenging process, but with patience and perseverance, you can achieve your desired outcome.
Below are some tips that may help you in your creative journey:
• Set a clear goal: Before you start creating, it’s important to have a clear idea of what you want to achieve. This will help you stay focused and motivated throughout the process.
• Experiment with different techniques: Don’t be afraid to try new things! Experimenting with different techniques can help you discover new ways of creating and lead to unexpected results.
• Take breaks: Creating can be mentally exhausting, so it’s important to take breaks when you need them. This will help you stay refreshed and avoid burnout.
• Stay positive: Creating can be frustrating at times, but it’s important to stay positive and keep pushing forward. Remember that every mistake is an opportunity to learn and grow.

- - - - Original Post - - - - (Posted on: Monday, October 23, 2023)
My favorite genre of video games is strategy, which includes card games, puzzles, and role-playing games. Recently, I have been playing some role-playing games that let me customize my character with my own image or artwork. I always choose the rogue or ranger class in RPGs, but this time I decided to try something different: a Soul Blade rogue. This is a character who creates psionic blades with their mind, similar to Psylocke from the X-Men comics. To make a realistic profile image for my character, I used an AI tool that can generate graphic art. My character is a female half-elf with short red hair in a pixie cut and glowing green eyes. Her eyes match the color of her psionic blades. She wears leather armor and has sigils and fae symbols tattooed on her arms.

This image was detailed, but it did not match my specifications. AI often overlooked some aspects of the prompt, and some times it requires more resources to generate a closer approximation. In other words, "Ya gotta spend more to get more". Negatives: The character’s hair was too long, there was no indication of half-elf ancestry, and nothing glowed in the image.

This one looked better, had more of the details I was asking for, but still felt static. Pretty much a posed portrait style - I wanted more dynamic action in the image. Hair was shorte and redder. Blade glows, eyes not so much. No tats due to long sleeve. Back to the AI drawing board.

This image is lively and fun, but a bit too cartoonish; and like so many female characters generated using AI, big breasted. Her eyes are attractive, but they are green instead of the color I wanted. The sword is also red and fiery, not purple and glowing. This is a common problem with AI; it overuses the colors that are specified in the prompt. If I ask for green eyes, it also added green clothing and a green shaded background. I mentioned red hair, it made the sword a fiery red (not to mention the left hand's glove too.) It might be very hard to get the exact colors I want from the AI – but I won’t give up.

I wanted a more rogue-like outfit, so I asked for a hooded cape and got this. I’m happy with this look, even though it’s not exactly what I had in mind. It suits my character well. This image has taken me off the main goal, but I am liking this different look for a rogue character, (not a Soulblade, but still a Rogue.) I'm going to request several more images, get better views of the character’s face for the game's dialogues.

This second image zooms in closer, but the character in not identical to the previous image. No mark on the face, different hood, and of course different weapon. These are details AI cannot duplicate... yet. These are differences, but they are very minor. This image looks great to me. But the key difference is the change of camera's angle; by adding the phrase "from above" this gives the image a more dramatic visual than the typical eye level approach.

The last image is a zoomed-in close-up on her face. Love the lighting and eye focus. I'm saving these three Rogue images, transferring them to the game's default portraits folder in the game so these will be available if I want to play a standard Rogue character. I'll be sure to go back to generating the Soulblade character I envisioned, and will update if I succeed.
Hazzah!

Femme Fatale: Queens of Music

I encountered a prompt for Ai to generate images in a pulp magazine style mixed with some Frazetta tones. I started doing famous actors, which often had great results. But these images weren't original, nothing we hadn't seen before. Then, while listening to 80's music, I generated an image using Debbie Harry of the famed 80's band Blondie. I have always found her unusually attractive; something about her shy-like yet dominating quirky manner, style, and beauty.

I loved the way this came out. But a square-shaped frame would take away from any full-body action pose. A vertical portrait format would permit more details, background, and dynamic poses.

From here, I started switching to other dominant female singers throughout the past 60 years. Singers from Country to Pop to Rock to Blues to Indy music.
See if you can guess them all:

The image above, which depicts Lady Gaga, showcases significant enhancements in detail, shading, and likeness due to a recent upgrade in the AI program. This improvement demonstrates the continual evolution of AI in producing more accurate and detailed representations. Additionally, it highlights that the same prompt now yields substantially different results compared to previous versions due to these upgrades. It's important to note that some AI generators offer the option to revert to earlier versions of their software, allowing users to create images with similar styles using the exact same prompts, even after updates.