Inspiration
I love memes because they are short and funny, but mostly because they make my wife laugh.
But the issue is that I'm bad at making them. I have the ideas, but it takes a lot of effort to generate them, especially since the online editors have terrible UIs.
What it does
JustMeme basically creates the meme for you. You upload a picture and say what you're thinking, and it figures out the best way to say it and place the text.
How we built it
The Grid Trick: Before sending your picture to Gemini 3.0, JustMeme draws a blue grid over it with numbers. It looks like a map. This helps the AI "read" where the objects are instead of just guessing.
Challenges we ran into
Even with the grid, the AI sometimes got too excited. It tried to write text so big it spilled off the screen, which clipped it on export. I fixed this by changing how I talk to the AI. I updated the system prompt to prefer smaller font sizes (scale 0.4 to 0.7).
Accomplishments that we're proud of
I'm proud that I'm actually "dogfooding" it. It is also super lightweight. It feels like a real, usable product, not just a hacked-together demo. I even added "prebaked" memes so anyone can try it instantly without waiting. But mostly, I'm happy I have a meme generator that makes my wife laugh (happy wife = happy life)!
What we learned
I learned that it’s smart to use different models for different jobs. For example, I use gemini-2.5-flash-lite for generating quick ideas because it’s blazing fast, while the bigger models handle the heavy thinking.
Also, planning matters. If I let the AI start coding without a plan, it digs itself into a hole. But if I discuss the goal, architecture, and UI/UX first, it builds exactly what I envisioned.
What's next for JustMeme
If I can get the AI to put text on a meme without covering a face, it can be applied for more serious matters too: Perfect Slides, Smart Subtitles, Instant Ads, Tweaking Websites
Built With
- aistudio
- gemini
- react-18
- typescript
- vite
Log in or sign up for Devpost to join the conversation.