> ## Documentation Index
> Fetch the complete documentation index at: https://docs.rawchunks.space/llms.txt
> Use this file to discover all available pages before exploring further.

# Songwriter Agent

> How text, images and recordings fill the Studio form.

Songwriter prepares a complete musical draft: title, production prompt, vocal or instrumental choice, language, style tags, BPM, key, scale, duration and tagged lyrics. Cover artwork runs alongside that work. You review the draft in **CONTROL**, then generate audio separately.

![Studio Control during Songwriter preparation, with draft fields and a lyric-writing progress banner.](https://mintcdn.com/rawchunks/ma13LU4xVjl9w_v5/images/app/studio-songwriter.png?fit=max\&auto=format\&n=ma13LU4xVjl9w_v5\&q=85\&s=594bfe7fc3a39144613335891a99ddca)

Songwriter fills the current Studio form as fields and lyrics arrive.

## Combine the inputs

Text supplies your request directly. Nemotron's vision model turns an image into a musical brief. Audio supplies transcribed words and signal features. Image and audio preparation run in parallel when you provide both.

For audio, Rawchunks prepares mono WAV and measures pitch with YIN, tonal profiles, onset autocorrelation, groove, dynamics and texture. Cartesia transcribes first; ElevenLabs provides the fallback. A lexical filter removes hum, scat and event labels from the spoken brief. A recording can still guide the music through its signal features when it contains no words.

Studio records up to 30 seconds. The [Vibe walkthrough](https://docs.rawchunks.space/app/vibe.md) shows how to combine inputs.

## Read musical libraries and fill fields

Nemotron-3-Ultra uses three compact libraries: styles, parameters and Lyrics Canvas tags. Tool calls read the relevant library and stream setters for the Studio fields. Planning allows two turns with a 1,024-token budget per turn and temperature 0.6.

Vocal drafts then use a separate lyric stream with a 1,200-token budget and temperature 0.75. Structural and performance tags stay in English while lyrics follow the vocal language. Instrumental drafts skip vocal lyric writing.

## Return the draft early

Nebius Token Factory serves planning, lyrics and vision with streaming enabled and thinking disabled. Field updates reach Studio as tool calls finish; lyric deltas arrive as text. Vision uses a short 180-token output budget and temperature 0.4.

Artwork starts alongside the lyric stage. `songwriter_ready` releases the musical draft before artwork finishes; `done` closes the full run after the cover result. This overlap keeps artwork off the critical path for reviewing and generating the song.

Use the [Songwriter integration](https://docs.rawchunks.space/dev/guides/generate/vibe.md) for the event contract. [Models](https://docs.rawchunks.space/ai-ml/models.md) lists the exact model IDs.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.