ACE-Step 1.5 ranks #2 of 8 in our ai music models testing. mit-licensed, self-hosted, vocals in fifty languages, and a full song in under ten seconds on a 3090..
88/100
the most permissive licence in the category and the only one you can audit line by line — with no vendor to hold responsible for anything.
why people look for an alternative
−no terms of service, so output ownership is unaddressed
−training data undisclosed
−no vendor support, uptime or moderation
−benchmark claim is the maintainers' own
stay with ACE-Step 1.5 if plain mit — no revenue cap, no attribution, no restrictions is the thing you care about most — nothing below beats it on that.
#1 in ai music models · the only one that licensed its training data before launching instead of after being sued.
91/100
verdictthe cleanest legal position in the category by a wide margin, and you retain rights to what you make — the catch is what 'commercial use' excludes.
Eleven Music vs ACE-Step 1.5
ACE-Step 1.5
Eleven Music
price
free
$6/mo
free tier
yes
yes
price
free (self-hosted)
$6/mo for commercial
who owns output
unaddressed — mit covers weights
you retain rights
free tier commercial
permitted
not permitted
litigation
none identified
none against the product
stems
unconfirmed
unconfirmed
switch foranyone putting generated music into something they sell, who would rather not follow a court case.
pros
+trained on licensed data with merlin and kobalt deals signed pre-launch
+no copyright litigation against the product
+you retain rights, not stated to lapse on cancellation
+vocals via a curated library or your own reference
cons
−free tier is non-commercial only
−self-serve licence excludes film, tv and studio games
−elevenlabs takes a perpetual licence to your output for training
−music consumes 900 credits per minute from a shared pool
#3 in ai music models · two of the three majors have settled and signed deals — the lowest legal risk of the two big generators.
82/100
verdictthe settlements are the product here — universal and warner are now partners rather than plaintiffs — and udio's own pages are the hardest to read in the category.
Udio vs ACE-Step 1.5
ACE-Step 1.5
Udio
price
free
unverified
free tier
yes
yes
price
free (self-hosted)
not readable
who owns output
unaddressed — mit covers weights
user, per help centre
free tier commercial
permitted
permitted, with credit
litigation
none identified
umg + warner settled; sony live
stems
unconfirmed
unconfirmed
switch forpeople who want suno-class output with meaningfully less unresolved litigation attached.
pros
+universal and warner settled and signed licensing deals
+commercial use permitted on the free tier with attribution
+help centre states users own the output
+reported royalty and audit terms give the settlements substance
cons
−pricing and terms pages unreadable without javascript
−sony litigation still live, ruling slipped to 2027
#5 in ai music models · an open-weight model with a licence you can audit, attached to a hosted product whose terms you can't read.
74/100
verdictthe only vendor here publishing a real open-weight licence for music — and the only one where the hosted product's own terms were completely unreadable.
Stable Audio vs ACE-Step 1.5
ACE-Step 1.5
Stable Audio
price
free
free under $1m revenue
free tier
yes
yes
price
free (self-hosted)
free under $1m revenue
who owns output
unaddressed — mit covers weights
user, per community licence
free tier commercial
permitted
permitted under threshold
litigation
none identified
individual suit, jan 2026
stems
unconfirmed
unconfirmed
switch fordevelopers under the revenue threshold who want weights they can inspect and self-host.
pros
+genuine open-weight licence you can read and self-audit
+free commercial use below $1m annual revenue
+no major-label litigation identified
+self-hostable rather than dependent on a service
cons
−$1m revenue cliff forces an enterprise negotiation
−hosted product's pricing and terms entirely unreadable
−'powered by stability ai' attribution required
−a training-data lawsuit from an individual musician is live
#7 in ai music models · changed name and owner twice in a year, and its own terms page returns 403.
64/100
verdictgoogle's infrastructure and synthid watermarking behind a product that has been rebranded twice and whose rights documentation we could not open.
Google Flow Music (formerly Riffusion) vs ACE-Step 1.5
ACE-Step 1.5
Google Flow Music (formerly Riffusion)
price
free
free with a google account
free tier
yes
yes
price
free (self-hosted)
not readable
who owns output
unaddressed — mit covers weights
user, per general google terms
free tier commercial
permitted
unverified
litigation
none identified
none against this product
stems
unconfirmed
unconfirmed
switch forpeople already inside google's ai subscriptions who want watermarked output with provenance.
pros
+synthid watermarking for provenance, unique here
+google's general terms state you retain output ownership
+free to start with a google account
+built on deepmind's lyria 3
cons
−product terms page returns 403 — rights unverifiable
−renamed and re-owned twice in twelve months
−no readable price list
−commercial use and monetisation terms unconfirmed
#8 in ai music models · keeps the copyright in the track you generated, and forbids you from releasing it as a song.
58/100
verdicthonest and clearly written about being a sync-licensing service rather than a music generator you own the output of — which is also why it ranks last.
Beatoven.ai vs ACE-Step 1.5
ACE-Step 1.5
Beatoven.ai
price
free
unverified
free tier
yes
no
price
free (self-hosted)
not readable
who owns output
unaddressed — mit covers weights
beatoven, always
free tier commercial
permitted
sync-only, all tiers
litigation
none identified
none identified
stems
unconfirmed
referenced, tier unconfirmed
switch forbackground music inside your own video, podcast or game, where a sync licence is all you need.
pros
+clear, well-drafted sync-licensing terms
+perpetual licence per download, surviving cancellation
+no royalties owed on qualifying use
+no litigation identified
cons
−beatoven retains the copyright in your generated track
−distributing the track to spotify or apple music is prohibited
every tool on this page went through the same test as ACE-Step 1.5 — same tasks, same order, scored the same way. the comparison tables are the figures from that testing, not vendor spec sheets.