Banana Prompts

I wanted to share something I’ve been hacking on for the last few months. It’s a little project called Banana Prompts, and honestly, it exists solely because I have the organizational skills of a squirrel.

Back when I first started getting deep into generative AI, I was obsessed. I was generating hundreds of images a day—experimenting with lighting, camera angles, distinct artistic eras. It felt like magic. But about three months in, the magic started to feel like a chore.

The problem wasn't the tools; it was the debris.

My workflow was a disaster. I had thousands of images scattered across local folders, Discord servers, and a Google Drive that I was too scared to open. But the real pain point was the prompts. I had a text file on my desktop called "prompts_final_v2.txt" that was basically a graveyard of good ideas.

I remember distinctly spending an entire Saturday afternoon trying to recreate a specific "Neo-Noir Cyberpunk" look I had achieved weeks earlier. I knew I had the prompt somewhere, but after digging through endless Discord scroll-back, I realized it was gone. I had lost the recipe.

That was the moment I almost threw in the towel. It felt ridiculous to spend more time hunting for my work than creating it.

I looked around for a solution. I wanted something that didn't feel like a spreadsheet. I didn't want a complex SaaS dashboard with analytics and team management. I just wanted a gallery. I wanted to see my image, click a button to grab the prompt, and get back to creating.

Since it didn't exist the way I wanted it, I built it.

bananaprompts.fun is my attempt to tame the chaos. It’s a community gallery designed for what I call "Nano Banana Prompts."

The idea behind the "Nano" concept came from my own frustration with prompt bloat. I realized a lot of us are pasting in paragraphs of text when we really just need the core stylistic triggers. So I spent weeks curating and distilling prompts down to their most effective versions.

I’ve uploaded a bunch of my personal favorites to get the ball rolling. There’s a "NYC Eras Split Screen" style that took me ages to perfect—it blends different time periods in a single shot without confusing the model. There’s also a "Frozen Spell Sorcerer" prompt that finally nails that high-shutter-speed action look that AI usually struggles with.

I’m putting this out there today not because it’s a finished, polished product, but because I figure I can’t be the only one drowning in PNGs.

I’d love for you guys to take a look. It’s completely free. You can browse the styles, copy the prompts, or—if you’re brave enough to open your own "prompts.txt" file—upload some of your own discoveries.

I’m also genuinely curious: how are you guys handling this right now? Is everyone else just winging it with text files, or is there a secret Notion template I missed out on?

I uploaded my original, messy PRD on the site too, if you want to see the technical breakdown of how I’m handling the gallery backend.

Anyway, check it out at. Let me know if it helps you clean up your workflow or if I’m just projecting my own messiness onto everyone else.

Thanks for reading.

I've been messing around with the new Nano Banana Pro model since it dropped, and I think I've finally figured out why it's so finicky.

It seems like the reasoning engine is almost too strong. It overthinks simple prompts. I found that to get consistent results—especially for things like fisheye perspectives or specific camera f-stops—you have to be incredibly rigid with the syntax. You can't just talk to it like you do with Midjourney.

I realized I was wasting hours re-learning the same lessons every time I started a new project. So I decided to productize my own swipe file.

I built bananaprompts.fun as a static gallery for these "stabilized" prompts.

The goal was pretty simple: creating a resource where the prompts are pre-tested to work with Gemini 3’s specific quirks. For example, getting text to render correctly on a sign in a "futuristic cafe" takes a very specific structure that I’ve documented in the library.

I’m also using this project as a bit of an SEO experiment. I’m trying to see if I can capture long-tail traffic for specific art styles (like "lomo camera aesthetic" or "vine-bone transparency") by giving each prompt its own dedicated page.

I’m looking for feedback on the categorization. Right now I have it mixed between art styles and subject matter. Does that make sense to you guys? Or would you rather browse by "use case" (like assets, backgrounds, characters)?

Also, if you have a prompt style that consistently works for you, let me know. I’m trying to expand the "Typography" section because that seems to be the biggest pain point right now.

Thanks for taking a look.