ComfyRAM Manager

See your VRAM.
Keep ComfyUI flowing.

A small control panel for the memory ComfyUI is holding. Windows only, MIT licensed.

The problem

ComfyUI holds on to memory

  • Models stay loaded after a run, and the next one runs out of VRAM.
  • The usual fix is closing ComfyUI and losing the session.
  • Hunting for the right process by hand takes minutes you don't have mid-project.
How it works

Two triggers, one panel, ComfyUI's own free call

The panel finds your running ComfyUI by scanning for its port, so there is nothing to configure.

How ComfyRAM Manager frees VRAMTwo triggers, a hotkey and the Auto Guardian, call the panel, which asks ComfyUI to free memory and polls its status every four seconds.POST /FREESTATUS EVERY 4 SHOTKEYCtrl+Alt+Fworks from anywhereAUTOAuto GuardianVRAM over 85%, 15 s restPANELComfyRAM ManagerElectron panel + PowerShellRUNNINGComfyUIport found by scanningtrigger or targetthe paneldashed = status back
The panel

Five buttons, one hotkey

  • Free: asks ComfyUI to free VRAM, trims working sets, shows before and after
  • Clear RAM: system memory only, leaves the GPU alone
  • Detect: re-scans for ComfyUI's port after a restart
  • Security Scan: checks suspicious running processes
  • Kill All: last resort. It ends every Python process and container, ComfyUI included.

Ctrl+Alt+F frees VRAM from anywhere, mid-generation, without focusing the window.

Auto Guardian

On by default, quiet until it matters

  • Watches VRAM; at 85% it runs Free for you
  • Waits 15 seconds before it can fire again
  • Thresholds live in one CONFIG object, easy to change
  • Toggle it from the panel any time
Install

Under a minute

  • Download ComfyRAM Manager Setup 1.1.0.exe from Releases
  • Run it. Nothing to configure.
  • Start ComfyUI; the panel finds it on its own.
  • The window is 460 × 780 and polls every 4 seconds.
Get it

Open source, Windows only

  • MIT licensed, Electron front end, PowerShell underneath.
  • Kill All is destructive; save your work first.
1 / 7