Command Line Arguments

What are Command Line Arguments?

Command line arguments are additional settings and options specified when executing a program (here, ComfyUI's main.py).

Use this when you want to control more core parts such as VRAM usage and network connection than the normal settings screen.

How to Set

Start by adding arguments after python main.py.

python main.py --fast

In the portable version, you can set it by right-clicking and editing the .bat file used for startup (e.g. run_nvidia_gpu.bat) and adding arguments after python.exe.

---

List of Arguments

VRAM / Memory Control (Stability and Performance)

Used for countermeasures against errors due to lack of VRAM (OOM) and operation adjustment according to VRAM capacity.

	Argument Name	Function	Remarks
🔥	--reserve-vram [GB]	Reserve VRAM: Sets the amount of VRAM to use in GB units, and reserves that amount for the OS and other apps.	In the case of --reserve-vram 1.0, at least 1.0GB remains.
🔥	--disable-smart-memory	Forced release of memory management. Does not keep the model in VRAM and actively offloads it to RAM.	Effective when you want to prioritize stability in an environment with low VRAM. Speed may drop.
	--lowvram	Enable VRAM saving mode.	Splits and executes UNET partially to reduce VRAM usage. Standard setting when VRAM is low.
	--highvram	Keep the model in GPU memory without unloading it to CPU memory.	Improves model switching speed in environments with large VRAM capacity.
	--cpu	Use CPU for all processing.	Very slow, so use as a last resort when there is no GPU.
	--normalvram	Force normal VRAM usage even if lowvram is automatically enabled.

Startup / Connection Settings (Convenience and Network)

Used for browser behavior and ComfyUI sharing settings within LAN.

	Argument Name	Function	Remarks
🔥	--disable-auto-launch	Disable browser auto-launch.	Convenient when auto-launch is unnecessary, such as in the portable version.
🔥	--output-directory [path]	Set output directory.	When you want to change the save destination of generated images.
	--listen	Allow external connection: Specify the IP address to listen to.	Listen on 0.0.0.0 (all addresses) with no argument (--listen). Accessible from another PC.
	--port [num]	Set listen port (default: 8188).	Change when the port is already in use.
	--extra-model-paths-config [path]	Load one or more extra_model_paths.yaml files.	When you want to manage multiple model storage locations.
	--auto-launch	Automatically launch ComfyUI in the default browser.

Troubleshooting (Error Isolation / Debugging)

Used when ComfyUI stops starting or when defects such as black images appear.

	Argument Name	Function	Remarks
🔥	--disable-all-custom-nodes	Disable loading of all custom nodes.	Most useful for isolating whether a custom node is the cause when ComfyUI stops starting.
	--force-fp32	Force FP32 (single precision).	For troubleshooting precision issues such as black images appearing.
	--disable-xformers	Disable xformers (acceleration library).	Isolate when xformers causes errors or black images.
	--fp16-vae	Run VAE in FP16 (half precision).	Saves VRAM during VAE execution, but black images may be generated.
	--cpu-vae	Run VAE on CPU.	Last resort to leave VAE processing to CPU when VRAM is extremely low.
	--disable-metadata	Do not save prompt metadata to file.	When you want to reduce the image file size.
	--fast	Enable untested optimizations.	For speed testing. Quality is not guaranteed.

Others (Advanced / For Developers)

Mainly includes advanced tuning of VRAM/precision and frontend testing.

	Argument Name	Function	Remarks
	--cuda-device [ID]	Set the CUDA device ID used by this instance.	For multi-GPU environments.
🔥	--front-end-version [version]	Used to try the new UI (frontend).	Example: --front-end-version Comfy-Org/ComfyUI_frontend@latest
	--temp-directory [path]	Set ComfyUI temporary directory.
	--multi-user	Enable per-user storage.
	--verbose	Increase debug prints.	For detailed error tracking.
	--force-channels-last	Force channels-last format during model inference.