Seed-Music
AI music, in your browser.
Turn a lyric, a 10-second voice clip, or a reference melody into a finished song. Seed-Music is ByteDance's research-grade music model — now usable without a GPU, an install, or a paper to read first.
Real Seed-Music tracks, made this morning.
Real prompts, real Seed-Music output. Press play — no sign-up.
3 AM Study
“lo-fi beats, soft rain, jazzy piano, melancholy”
Bedroom Floor
“boom-bap with dusty samples, lo-fi drums”
Trailer Cut
“epic taiko, swelling strings, brass hits”
Drift Run
“memphis phonk, distorted 808s, cowbell”
Tokyo at Night
“retro analog synths, neon midnight drive”
Deep Focus
“slow evolving pads, gentle textures, no rhythm”
Summer Hit
“dreamy female pop vocal, airy chorus, sunny”
Boss Battle
“chiptune boss fight, pulse arpeggios, 16-bit”
What is Seed-Music?
Seed-Music is the music-generation framework published by ByteDance Research in September 2024 (arXiv:2409.09214). It combines three models in one system: an autoregressive language model that composes full songs, a diffusion model that edits individual notes inside a finished track, and a zero-shot singing voice conversion model that clones a singer from a 10-second sample. We run that same Seed-Music research stack and ship it as a browser product.
Lyrics-to-Song
Paste your lyrics and Seed-Music writes a full vocal track — consistent melody, structure, and style across verses, choruses, and bridges. The Seed-Music Lyrics2Song head solves the long-range drift problem that breaks most AI songs after 90 seconds.
Multi-modal reference input
Seed-Music accepts five input modalities — text descriptions, lyrics, reference audio, musical scores, and voice prompts. Mix any of them to shape your song. Most AI music tools accept only text.
Zero-shot voice cloning
Upload 10 seconds of any voice — singing or speech. Seed-Music's SVC module generates a full song in that voice. Consent checkbox and audible watermark on every clone.
Edit melody & chords visually
Seed-Music exposes the generated song as an editable lead sheet — chords, melody, and lyrics laid out like real notation. Change a chord, swap a note, regenerate only that bar.
Inpainting — edit without regenerating
Don't like one note or one lyric line? Click and rewrite it. Seed-Music's diffusion editor re-renders only the affected region. The rest of the track stays bit-identical.
Studio-grade 44.1kHz stereo
Every Seed-Music track exports as 44.1kHz/16-bit stereo WAV — broadcast quality. MP3 320kbps if you prefer smaller files. Stems available separately for mixing in your DAW.
Six ways creators ship with Seed-Music.
Indie artist demo & release
Sketch demos with Seed-Music before booking studio time. Replace a $500 studio session with a 60-second render. Drop the demo straight to your collaborator or release it as is.
YouTube & podcast background music
Seedmusic delivers royalty-free background music in under 60 seconds. No Content ID claims, no licensing fees — drop a Seed-Music track straight into your video or podcast.
Game & app soundtracks
Adaptive loops, ambient beds, combat stingers. Seedmusic exports stems for Unity and Unreal — score every scene without a 5-figure custom-music budget.
Brand & ad audio logos
Generate brand jingles, ad music, and audio logos with Seed-Music. Iterate 20 variations in the time one composer delivers one. Commercial license included on Pro.
Voice cover & SVC production
Upload 10 seconds of any voice. Seed-Music's zero-shot SVC clones the singer and generates covers in that voice. Consent and audible watermark required on every clone.
TikTok, Reels & Shorts
Quick-render Seed-Music tracks for TikTok, Reels, and Shorts. Trending-sound length, claim-free, ready to upload — generated in under a minute.
Seed-Music vs Suno, Udio, and Mureka.
Suno and Udio are mature consumer products. Mureka is the closest peer to Seed-Music. None of them ship multi-modal input, 10-second voice cloning, lead-sheet editing, and arXiv-published research all in one place.
| Capability | Seed-Music | Suno | Udio | Mureka |
|---|---|---|---|---|
| Multi-modal input (text + audio + score + voice) | Yes | No | Partial | Partial |
| Permanent free tier | 10 songs / mo | 50 credits / day, watermarked | Limited | Limited |
| 10-second zero-shot voice cloning | Yes | No | No | No |
| Lead-sheet & note-level editing | Yes | No | Partial | No |
| Stem export on entry paid plan | Yes | No | No | Partial |
| Commercial license starts at | $19 / mo (Pro) | $10 / mo (Pro) | Pro only | Paid only |
| 11+ vocal languages | Yes | Yes | Yes | Yes |
| Open REST API | Studio plan | No | No | No |
| Underlying model published (arXiv) | Yes | No | No | No |
Bottom line: if you need multi-modal input, voice cloning, note-level editing, and an open API on top of an arXiv-published model, Seed-Music is the only one that ships all of it.
Start free. Pay when you ship.
No card for free. Cancel anytime on paid.
- 10 tracks / month
- MP3 320kbps
- Personal use only
- —No stems
- —No vocals
- Unlimited tracks
- WAV + 4 stems
- Vocals (11 languages)
- Commercial license
- Priority queue
- All 47 genre packs
- Everything in Creator
- 5 team seats
- REST + Webhook API
- Custom genre training
- Dedicated support
- SOC2 + DPA
Seed-Music questions, answered.
Is Seed-Music free?+
Yes. The free Seed-Music tier gives you 10 full songs per month with no watermark and no credit card required. Paid Seedmusic plans add commercial rights, stems, voice cloning, and higher generation limits.
How is Seed-Music different from Suno?+
Seed-Music ships three things Suno does not: 10-second zero-shot voice cloning, note-level diffusion inpainting, and multi-modal input (text + audio + score + voice). Suno is also closed-source — Seed-Music's underlying research is public (arXiv:2409.09214).
What inputs does Seed-Music support?+
Seed-Music accepts five input modalities: plain text descriptions, lyrics, reference audio clips, musical scores, and 10-second voice prompts. You can combine any of them in a single Seed-Music generation.
How does Seed-Music's voice cloning work?+
Seed-Music uses zero-shot Singing Voice Conversion: upload a 10-second clip of any voice (singing or speech) and the model produces a full song in that voice. A consent checkbox and audible watermark are required on every Seed-Music clone.
Can I use Seed-Music songs commercially?+
Yes — on Pro and Studio plans. Every Seed-Music track on those plans includes a full royalty-free commercial license valid for YouTube monetization, podcasts, ads, films, and games. Unlike Suno, you keep commercial rights even after canceling.
What languages does Seed-Music sing in?+
Seed-Music handles 11+ vocal languages including English, Mandarin, Cantonese, Japanese, Korean, Spanish, French, German, Portuguese, Italian, and Russian — with native-sounding vocals in each.
Can I edit a Seed-Music track after generation?+
Yes. Click any lyric line or melody region inside the generated track and rewrite it. Seed-Music's diffusion inpainting editor regenerates only the affected segment — the rest of the song stays bit-identical.
What audio quality does Seed-Music output?+
44.1kHz / 16-bit stereo WAV on all paid Seed-Music plans — broadcast quality. MP3 320kbps if you prefer smaller files. Stems available on Pro for DAW mixing. Up to 5 minutes per track.
Does Seed-Music offer a public API?+
Yes, included on the Studio plan. The Seedmusic REST API lets you embed Seed-Music generation into your own product or automation. Suno and Udio have no public API as of mid-2026.
Are there watermarks on Seed-Music tracks?+
No watermark on paid Seed-Music plans. Free-tier outputs are also watermark-free for personal use. The only exception is voice-cloned audio, which carries an audible watermark for safety.
Is every Seed-Music track royalty-free?+
Yes — on Pro and Studio plans. Every Seed-Music track is royalty-free: no per-stream payments, no claims against your video, no licensing fees. Personal use is free; commercial use requires Pro or above.
How does Seed-Music handle copyright and training data?+
Seed-Music is trained on licensed datasets. Generated outputs are checked against the training corpus to avoid memorization. Voice clones require consent and carry watermarks. We delete user prompts and audio within 24 hours and never train on user content.
Does Seed-Music support NSFW content?+
No. Seed-Music filters NSFW prompts at generation time. Vocal cloning of public figures is also blocked. Use the support channel to report misuse.
How do I get started with Seed-Music?+
Click any prompt on this page to try Seed-Music free — no signup needed for the demo. To save tracks, sign up for the free Seed-Music plan (10 songs/month). Pro is $19/month for unlimited generations.