🚀 ComfyUI Installer v1.2 — Multiple Instances, GPU Selection and Improved Restarts

The Inteliweb AI ComfyUI Installer for Pinokio has been updated to version 1.2 with improved installation controls, multiple-instance support and more reliable ComfyUI-Manager restarts.

✨ What’s new in version 1.2

🔁 Improved restart support with comfy-cli

ComfyUI is now launched through comfy-cli 1.12.0, which supervises the ComfyUI process and allows ComfyUI-Manager to restart it without closing the Pinokio Start session.

This supports:

• Manual restarts from ComfyUI-Manager.
• Restarts after installing custom nodes.
• Automatic restarts after dependency installation.
• Multiple consecutive restarts.
• Stop and Start directly from Pinokio.

The installer disables comfy-cli telemetry internally.

---

🌐 Custom ComfyUI port

A new setting is available:

COMFY_PORT=8188

Keep port 8188 for a normal installation.

Change it only when you want to run multiple independent ComfyUI installations simultaneously:

Installation 1: COMFY_PORT=8188
Installation 2: COMFY_PORT=8189
Installation 3: COMFY_PORT=8190

Each running installation must use a different port.

---

🎮 Select a single GPU per installation

A new GPU selection option has been added:

COMFY_GPU_DEVICE=auto

Available values:

auto → use the default GPU
0    → use the first GPU
1    → use the second GPU

Each installation can use only one GPU. Values such as 0,1 are intentionally unsupported.

Two installations on the same GPU

Installation 1: COMFY_PORT=8188, COMFY_GPU_DEVICE=auto
Installation 2: COMFY_PORT=8189, COMFY_GPU_DEVICE=auto

Two installations on different GPUs

Installation 1: COMFY_PORT=8188, COMFY_GPU_DEVICE=0
Installation 2: COMFY_PORT=8189, COMFY_GPU_DEVICE=1

Running two instances on the same GPU is supported, but both processes will share the available VRAM.

---

🧠 Improved version configuration

The recommended ComfyUI version setting is now:

COMFY_VER=auto

This installs and tracks the newest available ComfyUI version.

A fixed release can also be installed when compatibility with an older workflow is required:

COMFY_VER=v0.28.0

The comfy-cli version remains pinned to the tested release:

COMFY_CLI_VER=1.12.0

Advanced users can use:

COMFY_CLI_VER=auto

to install the newest version available from PyPI.

---

⚡ Automatic PyTorch and GPU backend selection

The recommended setting remains:

TORCH_VARIANT=auto

Automatic behavior:

Windows NVIDIA → PyTorch 2.10.0 + CUDA 13.0
Linux NVIDIA   → PyTorch 2.10.0 + CUDA 13.0
Windows AMD    → DirectML
Linux AMD      → CPU fallback
Apple Silicon  → MPS
Intel Mac      → CPU

Manual options are also available for CUDA 12.8, CUDA 13.0, DirectML, ROCm and CPU.

---

🖥️ CPU mode

Computers without a supported GPU can use:

TORCH_VARIANT=cpu

CPU mode automatically ignores COMFY_GPU_DEVICE and starts ComfyUI with the appropriate CPU argument.

---

🧹 Cleaner installation process

The temporary ComfyUI smoke launch was removed from the installation process.

ComfyUI now starts only once through the definitive comfy-cli supervisor after the installation finishes, reducing installation time and avoiding unnecessary temporary processes.

The interface button was also renamed:

Open Web UI → Open ComfyUI

---

🔗 Updated official ComfyUI repository

New installations now clone ComfyUI from its current official repository:

https://github.com/Comfy-Org/ComfyUI.git

---

✅ Tested scenarios

Version 1.2 has been tested with:

Windows 11
NVIDIA RTX 5090 Laptop GPU
Python 3.12
PyTorch 2.10.0 + CUDA 13.0
ComfyUI 0.28.0
ComfyUI-Manager V3.41
comfy-cli 1.12.0

Validated tests include:

• Installation from scratch.
• Manual and automatic ComfyUI-Manager restarts.
• Multiple consecutive restarts.
• Stop and Start from Pinokio.
• Custom ports 8188, 8189 and 8190.
• Two ComfyUI installations running simultaneously.
• Automatic GPU selection.
• Explicit GPU 0 selection.
• Changing the port after installation.

---

📌 Recommended upgrade method

Because version 1.2 introduces important changes to the installer structure, a clean installation is recommended instead of migrating an older Pinokio installation.

Use the default configuration for most users:

COMFY_VER=auto
PYTHON_VER=3.12
TORCH_VARIANT=auto
COMFY_PORT=8188
COMFY_GPU_DEVICE=auto
COMFY_CLI_VER=1.12.0

Thanks to everyone testing and improving the Inteliweb AI ComfyUI Installer for Pinokio.

Built with ❤️ by Inteliweb AI