GitHub

Quickstart

Install a release with one command, then open the studio in a browser.

Prerequisites

Check these on the machine before you install.

  • Linux on x86_64. The release has no ARM images, and it does not provide CPU inference or Intel or Apple GPU support.
  • Docker with Compose v2.27.2 or newer.
  • curl, tar, sha256sum, openssl, flock (from util-linux), and GNU coreutils (cksum, stat, readlink).
  • One of these hosts:
    • An NVIDIA GPU with the NVIDIA Container Toolkit (profile gpu).
    • A supported AMD GPU with /dev/kfd and /dev/dri (profile rocm).
    • No local GPU. This uses profile api, and a GPU worker on another host does the generation.

To check your machine against this list, see Checking a machine.

Install

Run this one command:

bash
curl -fsSL https://github.com/Portocolom-Studio/potocolom/releases/latest/download/install.sh | bash

The installer does these steps:

  • It checks the downloaded bundle against the release's SHA256SUMS file.
  • It installs into ~/potocolom. Set POTOCOLOM_DIR to use another directory.
  • It looks for an NVIDIA GPU first, then an AMD GPU. With no GPU, it selects the api profile. Set POTOCOLOM_PROFILE to gpu, rocm or api to choose yourself.
  • It pulls the released images and starts the stack.
  • It waits up to 120 seconds for the local API and store to pass /api/v1/ready. It does not wait for GPU warmup or model downloads.
  • If the API is not ready, it stops with an error and prints a logs command. Your data stays in place. Fix the cause and run the script again.
  • It prints a one-use setup link. On a desktop session, it also opens the link in your browser.

Choose how the install works

A new install starts unclaimed. Open the setup link that the installer printed. It has this form:

text
http://localhost:8080/setup#<token>

The link is good for one hour and works once. Choose one of these:

  • An administrator account, with an email and a password.
  • No accounts. Every request then uses one local administrator. This suits a single operator on a trusted network.

Keep the setup link private until you have spent it. If it has expired or was lost, see First run for how to get a new one.

Open the studio

Until you spend the setup link, the studio shows the setup page. After that, open the studio in a browser:

text
http://localhost:8080

The API container serves the studio.

Install models

A new install has no models. As an administrator, open Administration, then Models (/app?view=admin&tab=models). An empty model list in the Generate panel links there too. Each card shows the download size, whether the model fits your GPU, and a Suggested badge on the pick for it. Select models and download them. The worker refuses a download that would leave less than 5 GiB free, shows progress, and lets you cancel, retry or remove. Downloads are 2-7 GB for the SD and SDXL class models. A generation never downloads weights, so only installed models are offered.

Next steps

  • First run: the setup link, the choice, and what the studio does until you make it.
  • Installing a release: pin a specific release, and read the full install notes.
  • Reaching the studio: open the studio from other machines by setting the public URL.