Taan.ai
Taan.ai is a music model company building foundation models for music that follow user instructions in video.
- San Francisco, CA, USUnited States
Taan.ai is a music model company building foundation models for music that follow user instructions in video.
Which advantages over its rivals hold up, and why are they hard to copy?
Timing control holds up as Taan.ai's advantage over its rivals while measured audio fidelity does not, because Google Lyria has the lowest reported FAD: on its published timing benchmark against four commercial systems, Taan scored 80.1% instruction success and 73.2% on timed events within 500 milliseconds. The comparison rests on controlled 30-second instrumental requests. Its scoring gates only reject false matches and never adjust a detection to the requested time.
Taan.ai trains on fully licensed music produced with artists, and its output carries perpetual commercial-use rights. A revision touches only named parts, instruments, or timings instead of regenerating the track.
Where do rivals beat it, and what does each gap cost?
Taan.ai's timing lead does not extend to audio fidelity, where Google Lyria holds the lowest reported FAD. The benchmark uses controlled 30-second instrumental requests. The current preview is tuned for orchestral and cinematic music, with more genres planned.
Exported MP4 files omit the original audio track. The MCP route defaults to MP3 output, so a WAV file must be requested before generation or refinement.
Who does it compete with for the same buyer, and how do they differ?
The comparison is split by measure: TimingBench reports results for Taan and four commercial systems, Taan leads on timing, and Google Lyria leads on FAD. A user uploads a video and Taan creates an original soundtrack synced to timing and mood.
Taan connects to video tools and agents via an API and an MCP server, while its benchmark includes only short instrumental clips.
What does the product do for its buyer, and how is it delivered?
Taan produces an original score aligned with a video's pacing and emotional tone, constructed from video, script, and reference audio used together as inputs. The system examines the footage's cuts and mood to write the music in one pass. A requested change modifies only the specified parts, instruments, or timing points instead of regenerating the whole track.
Video tools and agents access Taan through an API and an MCP server, allowing them to call it directly. Outputs include a full mix and separate instrument stems.
