Fetching from the wire…
Public story · 2026-09-12 · high
Users still own generated tracks on every tier, including free, and ownership survives cancellation or a downgrade.
Why now: ElevenLabs published the change on September 11, making v2.5 the model most users now hit by default.
ElevenLabs made Music v2.5 the default generator for both prompted and reference-based tracks on September 11, with the older v2 model still available as a fallback. The company says the update produces richer melodies and more natural instrument timbres.
The rights terms change how people use it more than the audio claim does. Users own the tracks they generate on every tier, including free, and that ownership survives a cancellation or a downgrade, according to ElevenLabs' announcement. Free accounts get five lossless downloads a day. Pro accounts get 400 a month.
There's a catch. ElevenLabs blocks downloads of any track that references another artist's song, whether the reference comes from a text prompt or an uploaded audio file. The tool still generates that track. It won't leave the platform, though. The announcement also names a multi-year agreement with Universal Music Group, and the download block reads like the condition that agreement bought.
So the ownership promise has an edge to it. A free user can prompt a song 'in the style of' an artist signed to UMG and get audio back, but never be allowed to save it. ElevenLabs doesn't say whether the block applies to tracks made under v2, or how it decides a prompt 'references' a specific song closely enough to trigger it.
Each link below shares sources, entities, or timing with this story.
v0.5.1, tagged August 28, sits at 12,438 stars against only 23 open issues, which is an unusually clean ratio at that size. The AGPL-3.0 license is the constraint to notice: commercial embedding needs a separate arrangement, which is presumably deliberate positioning against E...
Stability closed a $76 million Series B on August 25 backed by Universal Music Group, Sony Music Group, Warner Music Group, Electronic Arts, AMD Ventures and Pacific Alliance Ventures, bringing total fundraising to $232 million. The investor list is the story. Three major labe...
TechCrunch reports the seed led by Coreline Ventures and Capital Today, with 8 million users at the one-year mark. The origin is the pattern: co-founder Shijia Liao, a former NVIDIA video researcher, trained the initial models on a single gaming GPU, and Fish Speech passed 31,...
Finally, a story about building something instead of worrying about something. Mistral released Voxtral TTS on March 26, an open-source text-to-speech model built on Ministral 3B. The numbers are striking: 90ms time-to-first-audio, 6x real-time factor (a 10-second clip generat...
SaaStr published Carles Reina's account of $0 to $600M+ ARR in 41 months, broken into $0-100M in 20 months, then $100-200M in 10, $200-330M in 5, and $330M-600M+ in 6. Quota is set at 20x base salary with attainment averaging 167% per quarter. POCs pay no commission whether th...
A model-routing API covering OpenAI, Anthropic, DeepSeek, Moonshot, Minimax, Nvidia, xAI, and Z.ai, with strategies including preferring flex usage tiers, specifying up to three performance benchmarks for automatic selection, or sending only complex queries to expensive models...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.