Scrambled Eggs

When Paul McCartney woke up with the melody of Yesterday in his head, he didn’t have any words for it yet. So he sang “scrambled eggs, oh my baby how I love your legs” until the real lyric showed up. That’s a dummy lyric: words that sit nicely on the melody and mean nothing, so you can feel out the phrasing before you commit to anything.

As a musician, melodies come pretty easily for me, but coming up with lyrics is a whole different story. I used to ask ChatGPT for some dummy lyrics to at least have some words that I can mumble. I wanted to build something that will help me write dummy lyrics so I can figure out the rhythm and phrasing and care about the meaning later (or never, usually never).

Scrambled Eggs makes those for you. You pick a rhyme scheme, it builds a grid of lines and fills every slot with a real English word that fits the stress pattern and rhymes where it should. The result scans, it sings and it’s complete nonsense. You sing against it, find where the long vowels want to land, then write the real lyric later.

This was the one rule from the start. A lot of songwriters would never touch an AI lyric generator, so there’s nothing like that in here. It’s pure dictionary lookup: about 133,000 pronunciations from the CMU Pronouncing Dictionary, each one tagged with its syllable stresses and how common the word is. Every word on the page was picked because it fits a rhythm, not because something predicted what should come next.

How it picks words

Every slot in the grid has a stress pattern, like 01 for “guitar”. Stress is the hard rule: a word only goes into a slot if its stresses match. After that it filters out rare words, prefers more singable ones (open vowels like “ah” and “oh” score higher than “uh” and “ih”, long consonant clusters score lower) and picks randomly from the best couple dozen, so every reroll is different.

Rhymes are matched on everything from the last stressed vowel to the end of the word. You can choose how strict that is:

• perfect: the whole tail matches (guitar / star)
• slant: same vowels, similar final consonant (water / daughter)
• assonance: only the vowels match

If a line ends on a tricky word and nothing rhymes properly, it falls back to matching just the last syllable (morally / body), so you never end up with one line that doesn’t rhyme with the rest.

What you can do with it

• Reroll everything, one section or one word. The rhyme sound changes too, not just the words.
• Lock a word you like. Lock it exactly, lock only its sound (it rerolls to another word with the same rhyme) or lock its meaning, which feeds the word cloud.
• Word cloud: words associated with your meaning-locked words, pulled from Datamuse. Click one and it shows you every slot it fits into without breaking the rhyme.
• Shape lock: freeze a line’s structure and reroll the rest of the section to match it.
• Type your own words or whole lines, change syllables and stresses by hand, add sections, save shapes as templates.
• Save / Open your project as a file. Undo and redo work on everything.

It runs entirely in your browser. There are no accounts and no server, your lyrics stay on your machine. The only things that leave it are the locked words sent to Datamuse for the word cloud and anonymous page-view stats.

Building it

The first version was a Python engine with a small Flask server in front of it. It worked, but it needed a server running somewhere just to look words up in a dictionary. So I ported the whole engine to plain JavaScript and baked the dictionary into one JSON file (about 1.3 MB gzipped). Now the app is just static files: the browser loads the dictionary once, builds its rhyme and stress indexes in memory and does everything locally. No build step, no backend, it deploys on Vercel straight from GitHub.