SEPARATION
Run it on your own machine
Separation on this site runs on a rented GPU, which is what a separation token pays for. You can also run it yourself, for nothing.
It is the same code either way. The GPU and your computer both run the
separation package, with the same model and the same refinement pass, so the
stems come out the same. The difference is time: a GPU takes seconds, and a processor takes
minutes.
What you need
A Windows PC and uv, a Python runner. Install it by opening a terminal and running:
winget install astral-sh.uvNothing else. uv fetches Python, PyTorch, and the separation model itself the first time you separate anything. That first run downloads a few hundred megabytes and takes a while; every run after it skips straight to work.
Separating a track
- Download Dayglass Separate and unzip it anywhere.
- Drag an audio file onto Dayglass Separate.cmd.
- Choose how careful to be. Balanced is the sensible default; ultra doubles the work again.
- Wait. A four-minute track takes roughly five to fifteen minutes depending on your processor.
The stems land in a folder beside the track, named after it, holding up to six files:
drums, bass, vocals, guitar,
piano, and other. Instruments the track does not contain are left
out rather than written as near-silence.
Measuring them
- Open the analyzer and load the same track you separated.
- Choose MY OWN STEMS.
- Point it at the folder the separator wrote.
From there it behaves exactly as though the separation had happened here: the same measurements, the same patch, the same export. Your audio never leaves your machine on this path, so there is no upload, no challenge, no daily limit, and no token.
Which to use
| Here | Your machine | |
|---|---|---|
| Time for a four-minute track | Under a minute | 5 to 15 minutes |
| Cost | 1 free a day, then a token | Nothing |
| Your audio | Uploaded, deleted after 90 minutes | Never leaves |
| Setup | None | Install uv once |
| Track length | 10 minutes | No limit |
If something goes wrong
"This needs uv" — install it with the command above, then open a new terminal window. An open one will not see it.
The analyzer says it found no stems — point it at the folder holding the files, not at the folder above it. It looks for files named after the six stems.
It is taking a very long time — that is normal on a processor, and ultra mode is four times the work of fast. Try fast first to see it finish.
Anything else: support@dayglassaudio.com.