Sometimes Prompts don’t work as you thought they would. See below for the prompts used!

Understanding the Limitations of AI Prompts: When Expectations Meet Reality

In the rapidly evolving world of artificial intelligence and creative content generation, users often experiment with prompts to get the exact output they envision. However, it’s not uncommon for prompts to fall short of expectations, revealing the current limitations of AI models and prompt engineering. Below, we explore real-world examples of prompts used to generate vivid video concepts and how subtle differences in wording can lead to unexpected results.

Case Study: The Cryptid in the Supermarket Aisle

Consider the following prompts designed to generate a handheld, selfie-style video featuring a mysterious cryptid known as “Monthman.”

Prompt 1:

“Generate a handheld POV selfie-style video of a tall, mysterious, owl-eyed Monthman-like cryptid, covered in dark feathers with a pair of shimmering, calendar-shaped wings. He stands in the cleaning aisle of a brightly lit supermarket, gripping a selfie stick with his talon-like hand. The light is harsh fluorescent and casts sharp shadows among rows of cleaning supplies. Footage shakes gently as Monthman awkwardly navigates a wobbly shopping cart. You hear distant announcements, squeaking wheels, and the soft rustle of wings. Monthman looks into the camera, and he says in a slightly nervous, baritone voice with a hint of midwestern accent, ‘They say I only show up to warn about disasters, but today’s emergency is that I ran out of floor cleaner!'”

Prompt 2:

“Generate a handheld POV selfie-style video of a tall, mysterious, owl-eyed Monthman-like cryptid, covered in dark feathers with a pair of shimmering, calendar-shaped wings. He stands in the cleaning aisle of a brightly lit supermarket, gripping a selfie stick with his talon-like hand. The light is harsh fluorescent and casts sharp shadows among rows of cleaning supplies. Footage shakes gently as Monthman awkwardly navigates a wobbly shopping cart. You hear distant announcements, squeaking wheels, and the soft rustle of wings. Monthman looks into the camera, and he says in a slightly nervous, baritone voice with a hint of midwestern accent, ‘Any tips for a cryptid on a budget? Furthermore, why are there 87 types of mop?'”

Analyzing the Results

Despite the similarities, these prompts can produce markedly different outputs. This discrepancy arises from how AI models interpret and prioritize details in prompts. Subtle variations—