What it does
You type something like "sunset reggae with steel pan, 78 bpm" or "a slow soul song about 20 years married". Pick a length from one to four minutes and tap Make my song.
Song Forge writes lyrics that fit the idea, generates the full arrangement with sung vocals, and drops the finished song into your song box on the phone. You can play it, save it, and use it however you like.
The songs are rendered on Nice Dreamz's own Apple Silicon hardware, not on a big-tech AI cloud. No third-party AI API is involved.
Screenshots

One sentence in, a song out 
Start free 
Describe it 
Watch it forge 
No subscription 
The splash screen
Who it's for
- Anyone who wants a real song for a birthday, an anniversary, a team, or a joke between friends, without learning an instrument or music software.
- Video makers who need a track they own outright.
- People who would rather describe music in a sentence than build it on a timeline.
- Anyone who does not want another subscription or another account.
Key features
One line in, full song out
Name the genre, the instruments, the voice and what it's about. Example prompts are right there if you want a starting point.
Lyrics with guardrails
A local Gemma model writes the words. The server rejects output that contains banned stock phrases, retries, and falls back to generated seed lyrics if the lyric model is down.
Real vocals, any genre
ACE-Step makes the music and the singing. Song Forge adds per-genre vocal cues for gospel, soul, hip-hop, reggae and more, so every style does not come out as the same pop voice.
Lyrics that stay in sync
Whisper times what was actually sung, then the server fuzzy-matches it back to the written lyrics so the lyric timing follows the song without showing transcription mistakes.
No account, no tracking
The app makes a random ID on first launch to count your free songs. No sign-in, no ads, no ad SDKs, no third-party analytics.
Songs land on your phone
Finished songs are delivered to your device. On Android they save through the system download manager. Nice Dreamz does not keep copies of finished songs on its servers.
FAQ
How much does it cost?
The app is free and comes with 10 songs. On iPhone, 30 more songs are $2.99, billed once through your App Store account. There is no subscription.
On Android, buying more songs is not available yet. The free songs work today.
Do I need an account?
No. No sign-in, no password, no email.
Who owns the songs?
You do, with full commercial rights. Sell it, gift it, put it in your videos.
How long does a song take to make?
The app's own screen says a full-length song in about a minute. Actual time depends on the length you pick and how many songs are in the queue.
Where are the songs made?
On Nice Dreamz's own Apple Silicon machines, running open models: ACE-Step for music, Gemma for lyrics, Whisper for lyric timing, on Apple's MLX. Your typed description is sent there so the song can be made.
Is it accessible?
Making a song takes typing one sentence and tapping one button. No music skill, no timeline, no fine motor work on a piano roll. The Android app is portrait only.
Is it open source?
The iOS render server, web UI and Android client in the song-forge repo are MIT licensed. The engines it runs (ACE-Step, seed-vc, Whisper, Gemma, MLX) are upstream projects by their own authors, listed in CREDITS.md.
Privacy, in short
Two things leave your phone: the song description you type, and a random app ID made on first launch so the app can count free and purchased songs. That's it.
- No account, no name, no email.
- No location, contacts, photos, camera or microphone access.
- No ads, no trackers, no third-party analytics. Data is not sold or shared.
- Purchases are handled by Apple or Google. Card details are never seen.
- Delete the app and the random ID and your local song box go with it.
- For general audiences, 13 and up.
Read the full privacy policy (effective July 14, 2026).
Built on
Song Forge's server, lyric pipeline, lyric timing, voice swap pipeline, UI and Android client were built by Matt Macosko. The models and engines underneath are other people's work:
ACE-Step (music), seed-vc by Plachtaa (voice conversion), Whisper by OpenAI (timing), Gemma by Google DeepMind (lyrics, served through LM Studio), MLX by Apple's ml-explore team, Demucs, ffmpeg and Three.js.