If you’ve ever tried generating AI images for actual work, you already know the truth: it’s not magic. It’s a tug-of-war between what’s in your head and what the tool thinks you mean. Some days you type one line and get a perfect result. Most days… you get a weird hand, a floating ear and a background that looks like it was designed inside a dream someone forgot.
After spending way too many hours playing with prompts, I stopped blaming the tools and started fixing the process. Over the past year, I’ve been using both ChatGPT and Gemini regularly, and I figured out a few simple hacks that dramatically improved my results, without paying for fancy add-ons or relying on luck. Here are the five things that actually worked for me.
1) Use ChatGPT To Write The Prompt
This was my first breakthrough.
Most people open an image generator and immediately start typing vague stuff like “make a cool thumbnail” or “create a cover image for this topic” Then they wonder why the output looks random.
Now I use ChatGPT as my prompt writer first. Not because Gemini can’t write prompts, but because ChatGPT is ridiculously good at structuring messy ideas into clear instructions.
Instead of dumping one line, I ask ChatGPT to create a full prompt with:
Scene description
lighting + camera style
subject expression + mood
background details
realism level
composition for thumbnails (space for text, focus area, etc.)
It feels like cheating, honestly. You get a detailed, clean prompt that already sounds like something a designer would write.
And once you have that base prompt, the results become far more predictable.
2) Always Use “Reference Photos” (Yes, Even For Simple Images)
This is the hack most people ignore because it sounds optional.
But reference photos are a game-changer, especially if you’re trying to create images that look real, clean, and news-friendly. When you give the AI a visual anchor, it stops guessing so much. It understands the vibe better.
I use reference photos in both ChatGPT and Gemini, especially for:
scam thumbnails
tech explainers
serious facial expressions
desk setups and newsroom environments
smartphone shots
Even if the photo is just a basic “writer desk setup” image from your gallery or a stock-style reference, it makes the output more grounded and less “AI-artsy.”
3) On Gemini, Always Pick Pro Mode (Fast Is Quick, Pro Is Worth It)
This is where Gemini’s personality shows.
Fast mode is tempting because it’s quick. Great when you’re experimenting, testing angles, or generating options. But when I’m making an image I actually want to publish, I’ve learned something the hard way:
Gemini Pro gives better results. Period.
Yes, it takes time. Sometimes it feels like it’s thinking about life choices. But the output is usually cleaner, more coherent, and less error-prone.
For work images like scam alerts, security warnings, or professional thumbnails, Pro mode reduces the chances of weird mistakes. And that alone saves me time, because regenerating five “fast” versions is still slower than getting one good “pro” result.
4) Use Both Tools Together for the “Best Of Both Worlds”
This is my favourite trick and also the most practical one.
I stopped treating ChatGPT vs Gemini like a rivalry. For me, it’s more like having two teammates with different strengths.
My workflow often looks like this:
Step 1: Use ChatGPT to craft a strong, detailed prompt
Step 2: Generate the image on Gemini Pro
Step 3: If Gemini gives something close-but-not-perfect, I tweak the prompt again in ChatGPT
Step 4: Run the improved version again on Gemini
Step 5: If needed, generate on both and pick the best result
Sometimes ChatGPT nails composition better. Sometimes Gemini wins on realism. Instead of forcing one to do everything, I let them do what they’re best at.
5) Rewrite The Prompt Instead Of Regenerating The Same One
This sounds basic, but it’s honestly the difference between “good enough” and “wow, this looks legit.”
Earlier, I would regenerate the same prompt again and again, hoping the AI would magically improve. That rarely works.
Now if I don’t like the image, I change the prompt. Even small changes matter:
“realistic newsroom lighting” instead of “cinematic lighting”
“serious, informative tone” instead of “dramatic warning vibe”
“natural face proportions” (yes, I literally add that sometimes)
“clean background with depth blur” for better focus
The moment you start treating prompts like a draft that needs editing, the AI starts behaving like a tool instead of a lottery machine.
