Sharpen your lyric before Suno hears it.
8 voices, 30 seconds, no signup.
Paste any draft. The room will name every line that's pulling its weight — and every line that isn't.
The words every model reaches for
Flagged in nearly every Crucible run on AI-assisted drafts. If your lyric leans on any of these, the line probably isn't earning its place yet.
Weak lyrics are a measurable pattern, not a vibe.
Every AI model — including the human-trained ones — reaches for the same 200-word vocabulary when asked for emotional lyrics. "Neon rain dances with shadows beneath the surface" reads as poetry the first time and as Mad Libs the tenth. The 12-metric rubric catches the pattern; the 8 voices name the specific line. Same diagnostic any working A&R applies to a demo.
Suno can't tell. The synth voice patches over weak lines.
Generated audio masks weak lyrics — the voice patch is convincing enough that the listener thinks the song is OK. Then they read the lyrics in the YouTube description and the spell breaks. Better to find the weak lines before the audio than after.
The verdict is a permalink. Share it. Argue with it.
Every Crucible run produces a shareable URL with your verdict. If a voice flagged "neon-rain-soaked streets" and you disagree, paste the URL in any forum and let the debate happen on the actual lyric, not vibes.
How the Crucible works
What a Crucible verdict actually looks like
One representative sample — click to expand.
Show
What a Crucible verdict actually looks like
One representative sample — click to expand.
Each voice returns one declarative sentence about your lyric. Quoted lines link the critique to the line that earned it. Below: an actual sample, lightly edited for length.
- Prosodist — The 4-stress cluster on “flicker once before they die” lands the meter clean; the chorus repeats the same scan and starts to feel mechanical.
- Specificity — “fluorescents” earns the room; “tell my mother I’m fine” survives because it’s a specific lie, not a general one.
- Antagonist — The pre-chorus is too thin: 1 line of preparation can’t earn the chorus you’re trying to land.
- Theme — Two competing theses: “I keep the porch light on for nobody” says abandoned, the chorus says recovering. Pick one.
- … 4 more voices in your real verdict.
Your draft gets all 8 voices, with line-quoted critiques + a single Survived / Wounded / Collapsed verdict.
Free · no loginOptional — Set the tone register
Tells each voice which emotional posture to score against. Quality is not universal — register defines craft. Why? →
Continue exploring
Why this works on AI-assisted drafts.
AI music tools (Suno, Udio, Riffusion) handle the voice + melody brilliantly. The lyric layer is where the craft signal lives. The Crucible audits that layer with the same rubric a working A&R or producer would — concrete imagery, body-anchored specificity, hooks that earn their position. The diagnosis is the same one used on songs that landed on Billboard last year. The room treats every draft seriously.
— Todd, founder. I built the Crucible because every AI-assisted lyric I wrote needed a producer's ear before I committed it to audio. Now you have one too.