How to Use Demucs in Gemini
Demucs separates mixed music into stems for practice, remix preparation, and vocal extraction. Its six-source model adds guitar and piano to the usual four stems, though piano separation can contain substantial bleed and artifacts.
Prompts to try with Demucs in Gemini
Copy an example into Gemini, or ask in your own words.
Use Demucs to extract the vocals from the song I supplied. Give me the result and a short checklist for studying the singer's phrasing.
Use Demucs to separate the drums from this recording so I can practice the fills. Keep the original recording available for comparison.
Use Demucs to extract the bass from this song as a reference for learning the bass line by ear.
Use Demucs with the six-source model to extract guitar from this recording. Give me a checklist for comparing it with the original.
Use Demucs to separate this chorus twice, first with one random shift and then with four. Present the vocal results together so I can compare their sound.
How to set up Demucs in Gemini
Pick the way that suits you. Each takes about two minutes.
Use the Gemini connector
- Open Connected Apps
On your computer, go to gemini.google.com. At the bottom, click Settings › Connected Apps. If you don't see Connected Apps, click Personal Intelligence first, then Connected Apps.
- Add Spiral as a custom app
Under Custom apps, click Add a custom app and paste this MCP server URL:
https://api.spiral.ai/mcp
- Sign in to Spiral and approve
Follow the on-screen instructions. Spiral opens: sign in, pick the organization and click Authorize. No API key.
Use the Spiral skill
- Send Gemini this message
Your agent reads Spiral's skill and sets itself up.
set up https://spiral.ai/SKILL.md
- Sign in or paste a key
Approve Spiral in the browser, or give the agent a key from the console.
Use the Spiral CLI
- Install and sign in
Installs the Spiral CLI, signs you in and adds its skill to your coding agents.
npx -y spiral-cli@latest init
Call the Spiral API
- Create an API key
In the console, under API keys.
- Run Demucs
One request; the result comes back by polling or a webhook.
curl https://api.spiral.ai/v1/runs \ -H "Authorization: Bearer $SPIRAL_API_KEY" \ -H "Content-Type: application/json" \ -H "Idempotency-Key: $(uuidgen)" \ -d '{ "provider": "fal", "model": "fal-ai/demucs", "input": { "audio_url": "https://storage.googleapis.com/falserverless/model_tests/audio-understanding/Title_%20Running%20on%20Fal.mp3" } }'
Then paste one of the prompts above: Gemini finds Demucs on Spiral, shows the price and runs it.
Demucs providers and prices
Spiral runs Demucs through fal. Open a card for today's price, with every fee included.
Settings you can ask for
Demucs works from your audio URL. Ask Gemini for any of these in your prompt.
| Setting | Allowed values | Default | Description |
|---|---|---|---|
| Model | htdemucs, htdemucs_ft, htdemucs_6s, hdemucs_mmi, mdx, mdx_extra, mdx_q, or mdx_extra_q | htdemucs_6s | Demucs model to use for separation |
| Segment length | a whole number from 1 | Length in seconds of each segment for processing (minimum 1). Smaller values use less memory but may reduce quality. Default is model-specific. | |
| Shifts | 1 to 10 | 1 | Number of random shifts for equivariant stabilization. Higher values improve quality but increase processing time. |
| Overlap | 0 to 1 | 0.25 | Overlap between segments (0.0 to 1.0). Higher values may improve quality but increase processing time. |
| Output format | wav or mp3 | mp3 | Output audio format for the separated stems |
Every API field of Demucs Stem Separation is on its tool page: open its card above.
What can you do with Demucs in Gemini?
Demucs gives you separate musical parts to study from an existing recording. A chat app can help turn a separation request into focused practice material and a listening checklist.
Vocal practice
Extract the vocal stem from a song to study phrasing and entrances without the full accompaniment competing for attention.
Drum practice
Separate the drums for focused listening to fills, groove, and transitions before practicing the part.
Bass transcription preparation
Extract the bass stem as a listening reference while working out a bass line by ear.
Remix preparation
Separate a mixed recording into stems, then audition each stem before choosing material for an arrangement.
Guitar part study
Use the six-source model to extract a guitar reference for studying rhythm patterns and musical entrances.
Piano part exploration
Try the six-source model for a piano listening reference. Check carefully for other instruments bleeding through before relying on individual notes.
More on Demucs and Gemini
Overview
Demucs in other agents
Questions about Demucs in Gemini
Can Gemini use Demucs?
Yes. Add Spiral to Gemini as an MCP server, and Gemini can find and run Demucs like any built-in tool.
How much does Demucs cost in Gemini?
$0.00091 per second. You pay from your Spiral balance as you go, with no subscription, and a failed run is not charged. Your Gemini plan is separate.
Do I need a fal account to use Demucs in Gemini?
No. Spiral runs Demucs on its own fal account. You sign in to Spiral once and never paste a key.
How long does it take to set up Demucs in Gemini?
About two minutes: open Connected Apps, then add Spiral as a custom app, then sign in to Spiral and approve.
Does Gemini wait while Demucs runs?
Demucs runs in the background. Gemini starts the run, checks on it and gives you the audio when it's ready.
What is Demucs useful for?
Demucs separates mixed music into vocals, drums, bass, and other accompaniment. Those stems give you material for instrument practice, vocal study, and remix preparation.
Can Demucs isolate guitar and piano?
Yes, with the experimental six-source model. The developers warn that the piano stem can contain substantial bleed and artifacts, so audition it before using it in an edit.
Will Demucs produce perfectly isolated instruments?
Treat the results as estimated stems and check them by ear. Piano separation has a documented problem with other instruments leaking through and audible artifacts.
Does the 7.8-second limit restrict the length of my song?
That limit applies to processing segments in Hybrid Transformer Demucs. Demucs can split a recording into chunks and combine their predictions, so segment length is different from song length.
When should I increase random shifts?
Try more shifts when separation quality matters more than processing time. Demucs averages repeated predictions, so compare the result with your first pass before adopting the change.
Start using Demucs in Gemini
One Spiral account for every tool your agent needs. No subscription; you pay per run.
Paste this into Gemini:
set up https://spiral.ai/SKILL.md


