Make it a mess meme.
Most meme tools are stuck in 2009 - Impact font, two text boxes, a watermark nobody asked for. Build the one they should have made.
Memes are the internet's native language. The tools that make them are an embarrassment. Someone uploads a photo - their cat, their coworker mid-sentence, their toddler having a meltdown. A vision model looks at it and suggests six directions that fit that photo. They pick one, fine-tune it live, then send it out and watch the reactions land. Because a meme isn't finished until someone laughs.
Funny is the hard part. That's the whole point.
Six steps, working end to end. No fake screens, no dead buttons.
A photo comes in. Drag it, paste it, or snap one with the webcam.
A vision-capable LLM reads the image and returns six meme ideas - each a different format, each written for what's actually in the picture.
Six live previews built from the user's real photo. They choose one.
A canvas editor. Rewrite the caption, swap the template, change the font, nudge the text. Fast and tactile.
Export a clean PNG, or send a link. No signup, no app, no friction for whoever opens it.
Anyone with the link can react - laugh, vote, pile on - and the creator watches the reactions roll in live.
If it doesn't do these, it isn't done.
Accept a photo upload - bonus for clipboard paste or webcam.
Call a vision LLM that returns structured meme suggestions, not a wall of text.
Offer at least six distinct templates, built as reusable recipes - not static images.
Show live previews of every suggestion using the user's actual photo.
A canvas editor with editable, draggable text - proper outline, shadow, line wrapping.
Export to PNG and copy-to-clipboard.
Give every meme a shareable link, and let anyone react - no signup, no install.
Show the creator the reactions as they land, ideally live.
Work on a phone. Memes are made on phones.
Use any stack, any model, any API. We care that it works.
Use any stack, any model, any API. We don't care how. We care that it works.
Here's what earns points.
The suggestions are the soul of the product. Generic captions lose. Sharp, photo-aware, format-aware memes win. This is mostly about how well you prompt the model.😂
Does the editor feel tactile and fast? Does the exported meme look like something we'd actually post? Typography matters. Motion matters. Taste matters.🎨
A clean template system, sane state, graceful handling when the API is slow or fails. The share-and-react loop is real full-stack work - shareable state, persistence, live updates done cleanly are where you stand out.⚙️
The whole loop works on a phone, not just a 1440px desktop.📱
The detail that makes us go "oh, nice." Surprise us.✦
Worth chasing -
Anything you want. Any framework, any AI API, any libraries.
Encouraged. Use Claude, Cursor, whatever you'd reach for on the job. We're hiring people who ship, not people who memorize syntax.
Your submission has to be a live link. Vercel, Netlify, wherever. We click links, we don't clone repos.
A working prototype URL. That's it.
Drop the link below. Resubmit as many times as you like - we judge your latest.
↻ Latest submission wins. Ship early, ship often.