VIBE MUSIC SHARE
AI models & attribution
Effective September 9, 2026
Vibe Music Share uses MiniMax-Music3 and ACE-Step 1.5 XL Turbo for demo music. The personal playlist and custom studio use locally operated music models. Vibe Music Share is an independent service and is not endorsed by these model developers.
MiniMax-Music3
The model creates music from lyrics and musical direction. Our prototype runs downloaded model weights on an operator-controlled GPU computer. Using those local weights does not itself send your lyrics to MiniMax’s hosted API.
MiniMax-Music3 is provided under its Community License and Acceptable Use Policy. These include commercial attribution and safeguards requirements. We display its name in our studio and provide this disclosure alongside our Terms and Privacy Policy.
ACE-Step 1.5 XL Turbo
Some demo recordings use the official ACE-Step XL Turbo model. It generates music from a musical description, lyrics and timing information. We run its downloaded weights locally on the operator’s computer; this process does not send the lyrics to an ACE-Step hosted music service.
The ACE-Step weights are released under the MIT License. The included Qwen3 text encoder is released under Apache 2.0. Personal recordings retain their generation model in the owner’s saved recipe. Model comparison recordings are kept for internal evaluation.
Semantic search
We use Tencent WeMM-Embedding-2B in our Azure environment to index lyrics and musical descriptions for search. Text similarity does not establish acoustic similarity or prove that a generated song follows a requested key or genre.
Local writing assistant
Song-writing requests are configured to use Qwen3.8-27B on an operator-controlled computer reached through our private Tailscale network. It drafts original lyrics and musical directions from questionnaire answers and preference summaries. Mira, our personal AI music assistant, uses the same private Qwen server for conversation, lyric drafting and musical direction. Explicit song requests can be added to the private generation queue when the member enables that option. These workflows do not submit content to the model developers’ hosted APIs. Ratings adjust prompt preferences; they do not fine-tune the music model weights.
Other music generation models
We may add further local models or hosted providers. Before a new provider receives user content, we will identify it here, explain the processing and update the relevant privacy information. A recording’s model information reflects the model actually used.
What AI-generated music means
Outputs can contain mistakes, unexpected words, or similarities to other music. Generated recordings are not guaranteed to be unique or eligible for copyright protection. A fictional demo artist profile is not a real musician or a registered customer. Do not present generated vocals as a real person’s performance without permission.
Current availability
Demo songs are available to listen to and share. Member requests enter a private queue when the creation service is enabled. A local worker writes lyrics and renders songs; generation depends on that computer being available. Daily songs are optional and can be paused.