From Wonky Hands to Wonder: My Journey Through AI Image Generation

A little over a year ago, I, like many others, jumped into the exciting world of AI image generation. The promise was incredible: type a few words and watch a machine create art. My early experiments, using free tools back in the summer of 2023, were… interesting. The results were often strange, a bit messy, but always fascinating.

Today, I primarily use the AI tools within Canva to help generate ideas and bring concepts to life. The difference in quality and coherence in just over a year is stunning. Looking back at my creations, you can see a clear story of AI learning, improving, and still, occasionally, getting things wonderfully wrong.

The "Good Old Days" of AI Weirdness

To understand how far we've come, you have to see where we started. This image of two figures by a bonfire is one of my first-ever generations, and it’s a perfect time capsule.

At first glance, it’s a moody, atmospheric scene. But look closer, and the classic AI quirks of that era emerge:

The Uncanny Valley: The faces are smooth, emotionless, and doll-like. They lack the fine details that make a face feel human.
The Dreaded Hands: AI famously struggles with hands. Here, they are poorly defined blobs. The figure on the right holds a stick in a physically impossible way.
Dreamlike Smudging: The entire image has a slightly blurry, dreamlike quality, where objects don't feel fully solid in their environment.

This was the baseline—impressive in concept, but flawed in execution.

The Leap Forward: Coherence, Style, and… Gibberish

Fast forward to today. The images generated with tools like the one in Canva are worlds apart. They are stylistically consistent, sharp, and can produce incredibly cute and creative results.

Characters are now well-defined, vibrant, and full of personality. But that doesn't mean the AI is perfect. The mistakes have just become more subtle. You might see a sticker outline that fades away on one side, or notice that the lighting isn't quite right.

One of my favorite remaining "flaws" is how AI handles text. It understands the shape and flow of calligraphy, but it has no concept of letters or language. This beautiful ornament is a prime example—it’s decorated with elegant script that means absolutely nothing.

From Bugs to Features: When AI Gets Creatively Confused

Sometimes, the AI’s failure to perfectly understand a prompt leads to the most creative results. I asked for a Halloween scene with Minions in a graveyard, and this is what I got:

Instead of just placing Minions among the tombstones, the AI merged the concepts. We have tombstone-Minions, pumpkin-Minions, and ghost-Minions. Their goggle shapes are warped and their bodies melt into the scenery. It's a "mistake," but it's also more imaginative and artistically interesting than what I originally asked for.

The Guardrails of AI: What You Can't Create

As these tools have become more powerful, they've also become more restricted. To prevent misuse, AI image generators have a growing list of banned or filtered content. While this can vary by platform, the main categories are:

Not-Safe-For-Work (NSFW) Content: This is the most universal rule. Generating sexually explicit or graphically violent imagery is strictly forbidden. Prompts containing related keywords are immediately blocked.
Hate Speech and Symbols: You cannot create images that promote hatred, discrimination, or violence against individuals or groups based on race, religion, gender, sexual orientation, and other protected characteristics. This includes generating hateful symbols.
Depictions of Real People (Especially Public Figures): This is a major area of concern to prevent the creation of "deepfakes" and misinformation. Most platforms heavily restrict or ban generating photorealistic images of politicians and celebrities. Some are so strict that even trying to create an image of a generic "mail carrier" might get flagged if it looks too much like a real person.
Copyrighted Material: This is a tricky and evolving area. While you can still generate images "in the style of" famous artists, directly asking for copyrighted characters like Mickey Mouse or characters from a specific movie may be blocked. My Minion images show that this is still a gray area, but platforms are getting stricter to avoid legal issues.

A Final Gallery of some Glorious Glitches

Now I will show a few more pictures that were created in a weirdly charming way. These are the kinds of beautiful mistakes that make exploring AI so much fun.

The Warped Zebra
The zebra is adorable, but the AI has struggled with its famous stripes. On the zebra's hindquarters and back legs, the stripes become wavy, distorted, and don't wrap around the body in a natural way. They look more like a painted pattern than a natural animal coat. The flowers in the foreground are also somewhat generic and lack fine detail.

The Inconsistent Turtle
This character appears to be a pink, stylized version of the Pokémon Squirtle. It's generally well-done, but the AI's struggle with consistency is visible. The right foot is clearly defined with three toes, but the left foot is a less-defined shape. The light source also seems inconsistent, with highlights appearing in places that don't logically match.

The Halloween Cats

This image is a classic example of an AI focusing all its "effort" on the main subject while failing to properly render the background elements:

Foreground vs. Background Quality: The black-and-white cat sitting on the pumpkin is the hero of the image. It is detailed, has depth, and is drawn in a clear, specific style. The two white cats in the background, however, look like they were drawn by a completely different, less-skilled artist.
The Distorted White Cats: These two cats are the weirdest part of the image. They are very flat and two-dimensional. Their faces are misshapen, with strange, mask-like markings instead of proper feline features. Their bodies are simple, elongated shapes with little to no anatomical structure. They look more like bizarre cat-shaped decorations than actual characters in the scene.
Inconsistent Style: The detailed, expressive style of the main cat clashes heavily with the flat, distorted, and simplistic style of the background cats, making the whole image feel inconsistent and strange.

A Perfectly Imperfect AI Snowman

The Double Hat: The most obvious strange detail is the snowman's headwear. It isn't wearing one hat, but two, stacked directly on top of each other. There is a blue, wide-brimmed hat, and growing directly out of that is a pink Santa-style hat. This makes no logical sense and is a clear sign that the AI has combined two different ideas ("snowman with a hat") into one bizarre object.
Scarf or Cape? The snowman has a standard red-and-white striped scarf, but it also has a larger, flowing piece of pinkish-purple fabric draped over its shoulders like a cape or shawl. This garment seems to merge with the scarf and doesn't behave like real fabric would.
The Mystery Creature: To the left of the main snowman, next to the smaller one, sits a brown, lumpy creature. Is it a rock? A frog? A potato? It’s completely unidentifiable. This is a classic example of an AI "cryptid"—an object the AI generates to fill a space when it doesn't have a clear instruction, resulting in a strange hybrid blob.
Illogical Arms: The snowman's twig arms emerge from very strange places. The right arm seems to come out from under the cape, but the left arm pokes out from behind its head/neck area, which is anatomically impossible, even for a snowman.
Floating Ornaments: On the Christmas tree to the right, the ornaments hang in a slightly unnatural way. Some appear to be floating in front of the branches rather than hanging from them, and the strings are either invisible or impossibly thin, a common detail that AI often overlooks.