Jan 31 • General discussion
I Gave 6 AI Tools the Same Brutal Prompt. Only 2 Survived.
I'll be honest: I set these AI image generators up to fail.
The prompt was a beast—a split-screen "Expectation vs Reality" meme with specific text placement, a centered overlay, AND a footer. Think of it like asking someone to juggle while riding a unicycle... backwards.
Why the torture test? Because most people expect AI to nail complex requests in one shot. This experiment shows what really happens when you dump everything into a single prompt.
Remember: each model has strengths and weaknesses. These results only reflect how they handled THIS style of prompt—your mileage will vary.
THE CHALLENGE
Here's what I asked every AI to create:
Left Panel (Expectation): Clean futuristic UI, calm colors, labeled "AI CONTENT (IN THEORY)" with tidy text blocks and balanced composition.
Right Panel (Reality):Absolute chaos—overlapping text, crooked faces, random fonts, labeled "AI CONTENT (IN YOUR FEED)".
The Kicker: Big centered overlay text reading "THIS AIN'T IT" spanning both panels, plus a footer: "Learn to do AI right → The Prompt Lab"
Think of this as the Final Boss level of AI prompting. Four distinct text areas, precise layout control, and stylistic requirements all at once.
THE RESULTS: Who Survived?
🏆 GEMINI – Near Perfect Structure
What Worked: Gemini crushed the structural requirements. It nailed the "THIS AIN'T IT" overlay in a stylistically perfect 8-bit font and actually included the footer text—something most tools completely ignored.
What Missed: The "Reality" side showed cats instead of the requested "crooked faces and visual nonsense." Funny? Yes. But a bit too safe.
Pro Tip: Gemini loves structural hierarchy. Use clear "Left Panel / Right Panel" logic in your prompts.
🥈 DALL-E 3 (ChatGPT) – Text Rendering Champion
What Worked: DALL-E 3 is currently the king of readable text in images. The "Reality" side delivered genuine chaos with overlapping elements, and the footer appeared correctly.
What Missed: It got confused and duplicated "THIS AIN'T IT"—putting it at both the top AND bottom instead of centered. The "Theory" side also went too Sci-Fi when I wanted clean UI minimalism.
Pro Tip: With DALL-E, less is more. If you specify center overlay text, don't mention it elsewhere or the AI will "hallucinate" extra copies.
🥉 PERPLEXITY – The Meme Vibe Master
What Worked: That yellow "THIS AIN'T IT" was high-contrast and perfectly meme-worthy. The "Reality" side nailed the gibberish text chaos I requested.
What Missed: Completely ghosted the footer. The "Expectation" side looked like a generic server room stock photo, not the clean UI aesthetic I wanted.
Pro Tip: Perplexity (using Flux or SDXL) needs explicit aspect ratio commands and more detailed descriptions to avoid stock photo vibes.
CANVA – Template Trouble
What Worked: The top labels were perfectly legible and correctly positioned. If you need simple, clean text, Canva delivers.
What Missed: No central overlay, no footer, and the left side was literally just a blank template. This wasn't even close.
Pro Tip: Canva's Magic Media shines for single-subject generation. For complex layouts like this, generate the background separately and add text manually in Canva's editor.
MIDJOURNEY – Beautiful... But Wrong
What Worked: Image #1 is absolutely stunning. That cyborg head? Chef's kiss. Pure artistic brilliance.
What Missed: Everything I actually asked for. Midjourney struggles with specific layout instructions (left vs right panels) and long text strings. All text came out as AI gibberish like "EXPICATION TREENITY."
Pro Tip: Midjourney is for VIBE, not INSTRUCTIONS. To get this right, you'd need to use "Vary Region" or generate two separate images and stitch them in Photoshop.
CLAUDE – Perfect Logic, Zero Aesthetics
What Worked: Claude understood EXACTLY where every piece of text should go. The logic? 10/10.
What Missed: Claude created an SVG mockup instead of a cinematic image. The visuals? 1/10. That's because Claude is a language model, not a native image generator.
Pro Tip: Don't use Claude for "cinematic" images. Instead, use Claude to WRITE the perfect prompt for Midjourney or DALL-E.
THE VERDICT:
For a one-shot, ready-to-post result? Gemini or ChatGPT are your winners. They understood the Gen-X "This Ain't It" energy and actually delivered on the specific branding requirements.
But here's the real lesson: This prompt was intentionally brutal. Most AI tools need you to break complex requests into steps rather than dumping everything at once.
Want better results next time? Build your images in stages—background first, then text elements, then final touches. That's what separates casual AI users from people who get professional results.
Which AI do YOU think came closest? Drop your thoughts below. 👇
1
2 comments
I Gave 6 AI Tools the Same Brutal Prompt. Only 2 Survived.
powered by
skool.com/melissa-af-harmon-creative-9018
Built for the entrepreneur who's tired of winging it with AI.
Suggested communities
Powered by