AMSAMS — AI Music Studio
AMS user guide
Updated: 2026-11-01
Describe the music you want to hear
Write the genre, instruments, rhythm, and mood in Romanian or English. Genre buttons and suggestions append to your prompt without deleting your text. Manual BPM adds a model instruction; exact tempo is not guaranteed.
Prepare your device
Generation requires a browser WebGPU adapter. We check the adapter and storage when you press Generate. Models download from Hugging Face and can be cached locally. The full profile inventory is about 5.94 GB; instrumentals without the planner use a subset.
Wait for local processing
Generation uses your GPU through WebGPU; only final VAE decoding uses your CPU through WASM. The GTX 660 trial produced 10 seconds of audio in about 14 minutes. Other devices will differ. Download percentages use reported bytes; other operations are shown as stages.
Listen, edit, and download
The player shows a waveform from real audio, seeking, and volume. The editor provides local trimming, fades, gain, and normalization. Preview does not overwrite your library original. Download the WAV to keep a safe copy: browsers can remove local data.
Current version limits
Only the 10-second instrumental mode is qualified by a real trial. We do not promise generation on mobile, Safari, Firefox, or every GPU. Vocals, sound effects, MP3, and longer durations are not presented as available.
Local diagnostics: check your device and download the report