Connect your GPU PC

1. Prepare Windows

Update your NVIDIA Windows driver. Open PowerShell as administrator:

wsl --install -d Ubuntu-24.04
wsl --update

Restart if requested, open Ubuntu, create your Linux user and check the GPU:

nvidia-smi
sudo apt update
sudo apt install -y python3-venv python3-pip ffmpeg git curl unzip build-essential libsndfile1 libportaudio2

The GPU check should show your RTX 4080. WSL uses the Windows GPU driver.

2. Install the worker

Run these commands in Ubuntu. You will need the private worker token supplied with the deployment, separate from your app password.

cd ~
curl -fL https://take-two.kordlik.com/worker.zip -o take-two-worker.zip
unzip take-two-worker.zip
cd take-two
bash scripts/setup-worker.sh
bash scripts/start-worker.sh

Keep this terminal open. Audio files and saved takes stay in ~/take-two/data.

3. Open the connection

In a second Ubuntu terminal, install the official Cloudflare connector:

cd ~
curl -fL https://github.com/cloudflare/cloudflared/releases/latest/download/cloudflared-linux-amd64.deb -o cloudflared.deb
sudo dpkg -i cloudflared.deb
cloudflared tunnel --url http://127.0.0.1:8788

Copy its https://….trycloudflare.com URL into the GPU_ORIGIN secret in Cloudflare → Workers & Pages → take-two → Settings. The deployed WORKER_TOKEN must match the PC token.

Refresh Setup in the studio and run a synthetic audio check. It tests the connection, audio processing and downloads without performing voice conversion.

For a stable address, replace the quick tunnel with a named tunnel at an unused hostname such as take-two-gpu.kordlik.com. The full guide is included in your download.

4. Add a voice engine

Start with Seed-VC if you do not have a community RVC model. Stop the worker with Ctrl+C, then run:

cd ~/take-two
bash scripts/install-engines.sh seedvc
bash scripts/start-worker.sh

This installs the separator and Seed-VC in separate environments, checks CUDA, and updates the worker configuration. Allow several GB for dependencies and model downloads. The first preview downloads additional weights.

To add RVC, repeat with bash scripts/install-engines.sh rvc. Supply a compatible .pth voice model and its optional matching .index. Use trusted checkpoints; the app does not sandbox their loaders.

The installer is prepared from pinned upstream sources. Actual GPU conversion still needs its first test on your PC.

5. Preview, compare, keep

1. Upload your song and choose a representative 20-second excerpt.
2. Choose Seed-VC and upload a clean 5–20 second reference vocal, or select your RVC model.
3. Render a preview. Separation is reused on the next take.
4. Compare three parameter variants, inspect technical checks, and listen for voice quality.
5. Save a favorite and render the full song using its settings.

Keep Windows awake, the worker running, and the tunnel open while rendering or listening. Jobs continue if you close the browser.