Describe a song.
Get it in code.
Songcode turns a simple prompt into a song built out of code, with fully editable stems in our web DAW. There is no generative audio model (and there never will be).
Songcode uses AI to turn your prompt into playable code. There is no audio model (and there never will be).
Free with your own API key. We let new people in often. Have an invite? open songcode
Free with your own API key. what that means
odetitle "after the last train"tempo 120key c minorswing 55%form loop 4chords [Cm9 | Fm11 | % | Abmaj7 Gm7]let answer = line [r/4 c5/8 eb5 f5/4 eb5/8 c5 | bb4/4. g4/8 ab4/4 g4/4]stem kick: tight { kick [x... x... x... x...] velocity [112 100 106 98] decay 320ms bar 4 of 4 { kick [x... x... x... x.o.] }}stem rim: tight { rim [.... x..x .... x... | .... x..x ..o. x..o] velocity [96 70 90 64] highpass 300hz reverb 20%, tone: 4khz}stem hats: tight { hat [o.x. o.xo ..x. o.x. | o.x. ..xo o.x. ..xx] velocity [92 44 70 52 86 40 64] chance 92% highpass 500hz lowpass 9khz bar 4 of 4 { hat [o.x. o.xo ..x. o...] openhat [.... .... .... ..x.] }}stem shaker: cabasa { rhythm [xxxx xxxx xxxx xxxx] take [0 2 1 3 2] velocity [78 40 60 46 72 38 56] chance 85% highpass 900hz}stem bass: fmbass { line [held c2/4. g1/8' bb1/4 c2/8' eb2/8 | f1/4. c2/8' f1/4 ab1/8' c2/8 | f1/4. eb2/8' c2/4 ab1/8' g1/8 | ab1/4. eb2/8' g1/4 bb1/8' b1/8] velocity [108 72 90 76 84] lowpass sweep(450hz, 900hz, every: 4bars), resonance: 15% release 120ms}stem keys: epiano { rhythm [x--- ---. ..x- ---- | --.. x--- .... ..x- | ---- .... x--- .x-- | --.. x--- ..x- ----] tones [all] octave 3 voicing smooth velocity [74 58 68 62] highpass 220hz chorus 35%, rate: 0.6hz phaser 0.3hz, depth: 30% reverb 35%, tone: 4.5khz delay 18%, time: 3/16, feedback: 30%}stem pad: warmpad { rhythm [x] tones [all] octave 3 voicing wide attack 600ms release 1.8s highpass 300hz lowpass sweep(1.2khz, 2.6khz, every: 4bars) reverb 55%, decay: 5s, tone: 3.5khz}stem vibes: vibraphone { line [r/1 | r/1 | mp answer] pedal chord highpass 250hz reverb 35%, tone: 5khz delay 22%, time: 3/16, feedback: 35%}mix { kick 0db rim -10db, pan: 12L hats -11db, pan: 15R shaker -15db, pan: 25L bass -4db keys -8db, pan: 8L pad -16db vibes -11db, pan: 14R duck bass, keys, pad under kick, depth: 55%, recover: 140ms reverb decay: 2.2s, tone: 4.5khz master ceiling: -1db}This is the whole track. Press play, change a number, press update.
There is no audio model.
Most generative music comes from a model trained on recordings, and it hands you audio. Songcode works the other way. A general-purpose language model, the kind that writes software, writes a short program. Your browser turns it into sound.
Returns audio. You can play it.
stem kick: tight { kick [x... x... x... x...] velocity [112 100 106 98]}stem shaker: cabasa { rhythm [xxxx xxxx xxxx xxxx] take [0 2 1 3 2]}Returns music code. Immediately playable and editable in realtime.
We trained nothing
No dataset of songs, no fine-tune, no audio model of our own. We never trained on anyone's music.
The track is text
A loop is about a hundred lines. Every note and every filter setting is on the page, so you can edit it, diff it and keep it in git.
Sound is made in your browser
Synths, drum machines and effects run live in the page. No audio is rendered on a server.
A prompt goes in. Code comes out.
Write a prompt
Say what you want to hear. Pick a genre and a length.
Read the code
The model writes a song out of code. Every line works together to form a composition.
Change it while it plays
Edit a value and press update. The track keeps playing with your change.
Every part is a stem.
Switch from code to stems and the same song opens in our web DAW. This is the track from the top of the page. Mute a stem, solo one, or pick one to see the code that makes it.
stem bass: fmbass { line [held c2/4. g1/8' bb1/4 c2/8' eb2/8 | f1/4. c2/8' f1/4 ab1/8' c2/8 | f1/4. eb2/8' c2/4 ab1/8' g1/8 | ab1/4. eb2/8' g1/4 bb1/8' b1/8] velocity [108 72 90 76 84] lowpass sweep(450hz, 900hz, every: 4bars), resonance: 15% release 120ms}Mix it
Mute, solo, level and pan each stem. The mix is written into the code.
Edit the notes
A piano roll for pitched stems and a grid for drums. Draw, move and resize notes, and the code is rewritten to match.
Rewrite one stem
Ask for a different bassline and only the bass changes. It is written against the tempo, key and stems already there.
What you can do with a track.
60 genres
Eleven families, from techno and house to jazz, rock, ambient and solo piano. Each has its own tempo range and sounds.
Loop or full track
A seamless loop, or an arranged track of three to five minutes.
About thirty controls
Tempo, scale, drum kit, swing and more. Set what you care about. The model decides the rest.
Sounds of its own
24 drum kits, 53 instruments, six singers and nine ambience sounds built for Songcode, alongside recorded instruments from piano to electric guitar.
Source, WAV or MP3
Rendered in your browser, faster than real time. No audio is uploaded.
A link for every track
Publish a track to get a link. Anyone who opens it can play it, and anyone with a Songcode account can remix it.
Made for the tinkerers.
If you flash your own firmware onto a cheap synth, live-code, or would rather read the patch than trust the preset, Songcode is for you.
Free with your own key.
$0 a month
Bring an Anthropic or OpenAI key and Songcode charges nothing: no subscription and no fee per track. Your provider bills you for what the model uses, and we add nothing on top.
Or a plan coming soon
No key required. Songcode writes your tracks with models tuned for subscribers, from weekly usage, billed monthly.
Get early access.
Songcode is in early access, and we let new people in often. Leave your email and we'll send your invite.
Write a track.
An account, a prompt and your own key are all it takes. Songcode is free.
open songcode