How to Make a Song With AI (Lyrics, Vocals, Cover Art)

To make a song with AI you need three things: words, a description of how it should sound, and a generator that turns both into vocals and instruments. In ClipCraft's Audio Studio that whole loop happens in one browser tab. You write or paste a lyric sheet (or pay a 5-token bot to draft one), describe the style in plain words, pick a voice, and a few minutes later a finished song lands in your library with free cover art attached. A song costs 300 tokens, which I'll translate into real dollars near the end, because AI music pricing is usually where the fine print hides.

Step 1: the lyric sheet

Songs are structured with bracket tags: [Verse], [Chorus], [Bridge] and eleven more, fourteen in all, sitting in a fold-out under the lyrics box so you can tap one in at your cursor. The sheet caps at 3,500 characters, which fits a normal three-to-four-minute song with room to spare. If you're coming from Suno, paste your old sheets as-is: tags like [Verse 1] or [CHORUS] get converted automatically, with a small note showing what changed. Two rules I've learned to respect: keep every chorus identical (three different choruses stops sounding like a song), and leave [Solo] and [Inst] tags bare, or the vocalist will happily sing right over the guitar solo.

Don't want to write? The ✍️ Lyric bot lives one door over in the Lyrics room: describe the song in a sentence or two, say who's singing it (any, man, woman, or a duet that actually trades verses), pick a structure, and it writes the full tagged sheet for 5 tokens, charged only if it delivers something usable. There's a parental advisory switch if your song needs to swear, and a free blank-structure button if you just want the skeleton.

Step 2: style and voice

The style box wants the music described, not the story: genre first, then a few mood words, then one tempo or production cue. Something like "synthwave, moody, retro 80s, driving 104 BPM" beats a paragraph of vibes. A four-way picker chooses the voice (any, male, female, duet), and it works by quietly adding the right phrase to the front of your prompt rather than editing your text. If you already have lyrics, the ✨ let-lyrics-inspire button reads them and suggests a genre-plus-mood descriptor for 5 tokens, shown as a removable chip so it never eats your own words.

Step 3: make the song with AI, and what 300 tokens buys

Hit Create (or Ctrl+Enter) and the song composes for a few minutes, then lands in the library on the right side of the page as a full track with vocals, usually somewhere between 2:30 and 4:45, downloadable as 256 kbps MP3 or 44.1 kHz WAV. A "Why 300 tokens?" note under the button says the quiet part out loud: the GPU time is billed per file, the price runs a couple of times raw cost, and a failed run refunds automatically. Two details I lean on every time I use it myself: you can close the tab mid-generation and the run finishes anyway (it resumes polling next time you open Create), and every song gets free AI cover art generated from its title, no extra tokens, painted behind the track in your library. Our cover art post covers where else that art system shows up.

Want it to sound like something? Analyze the something

The 🔍 Song Analyzer is the studio's cheat sheet for style prompts. Drop any track on it and it detects tempo and key entirely in your browser (the page says "analyzed locally, nothing uploads" and means it), then hands you tap-what-you-hear chips for vocals, genre, mood, era and instruments that assemble into a Create-ready description. Honest receipt: I fed it a 120 BPM test loop I generated myself and it read 118, two taps of the +2 button from exact, and it called the key C major with the relative A minor listed right under it, which was a fair reading of an ambiguous progression. There's a tap-tempo button for verifying by ear, and an optional Whisper transcription (5 tokens per started minute) if you want the original lyrics in an editable sheet. It builds a style description, deliberately: it cannot and does not copy the melody.

After the song lands

• 🎚 Extract stems splits a finished track for 100 tokens: all six stems, or just vocals plus instrumental, which is also the one honest way to get a backing track of your AI song.
• 📥 Send to editor opens the song (or each stem) in SoundCraft, every file on its own channel, ready to mix like any recording.
• The library does playlists, likes and folders, with a player that keeps playing while you move between rooms; posting to the studio's feed or sharing it further is one dialog.

The money, plainly

A song is 300 tokens. The free plan's 100 monthly tokens cannot make one, full stop; free accounts are for the analyzer, the editors and the smaller tools. The $1.99 Saver plan includes 300 tokens a month, exactly one song; Creator at $7.99 covers six; and token packs start at $0.99 for 100 and never expire, so three small packs is a one-off song with no subscription. Compared with the subscription-only tools in our Suno alternatives roundup, the trade is simple: you pay per song instead of per month, and the file is a plain download you keep, with no watermark stamped into it. The account itself is free.

What to know before you spend tokens

• One song engine, no model menu; instrumentals have their own mode, and one-shot sounds get a whole SFX generator from 10 tokens.
• Songs top out around 4:45; there's no ten-minute epic setting.
• The analyzer describes; it doesn't clone. If you need the exact melody, that's what stems and a DAW are for.
• Vocal gender is prompt-steered, so a run can occasionally surprise you.

That's the full loop: words, style, voice, generate, stems, mix, post. The first one takes about ten minutes including signup; start free and bring lyrics.