Kivio
Home
Create
Model
Journal
Changelog
Create
Product
Changelog
Every GWEN model release and Kivio product update. Newest first.
Kivio Studio
Jul 2026 · live
Make a song, for real
GWEN is live.
The
Studio
now generates real songs on demand: type an idea, pick a genre and voice, and GWEN writes and sings it.
Your library.
Every song you make is saved to your
Library
, ready to play again or download.
It learns from you.
Rate a song and GWEN tunes the next one toward what listeners actually love, per genre.
GWEN v1.3
Jul 2026
Nearly 2× the training data
Vocal recovery.
Re-transcribed a large slice of the catalog with Whisper large-v3, recovering many real vocals the old pipeline had missed.
Broader coverage.
More Yoruba, Pidgin and English-Pidgin vocals folded in.
Cleaner pipeline.
Fixed a silent data-loss bug that had been dropping tracks during prep.
GWEN v1.2
Jul 2026
XL base, standalone model
ACE-Step XL.
Moved to the 4B XL base for a higher audio ceiling.
Standalone.
Merged the adapter into the base so GWEN ships as one self-contained model.
Bigger set.
Trained on the unified catalog across every source bucket.
GWEN v1.1
Jul 2026
Vocals that actually sing
Vocal-forward captions.
Rich, descriptive prompts so GWEN sings the lead instead of defaulting to instrumental.
Sharper mixes.
First pass at a mastering chain for radio-loudness output.
More genres.
Amapiano, gospel and highlife added to the core set.
GWEN v1.0
Jul 2026
First African vocal model
The foundation.
Fine-tuned ACE-Step on a consented African catalog.
Nine genres.
Afrobeats, Amapiano, Afro-house, Alté, street-pop, Afro-swing, gospel, highlife, Naija rap.
Catalog-sourced.
Built only on music we own or licensed.
Next
Roadmap
What we are building
Better decoder.
A new anti-aliased audio decoder to lift raw fidelity past the current ceiling.
African voices.
A voice library so you can pick the singer, built from consented artists.
Afrobeats genre model.
A dedicated Naija-afrobeats fine-tune for an even tighter, more authentic pocket.