How to config ComfyUI Startup Arguments on RunComfy

ComfyUI startup arguments (CLI args) are now fully exposed on RunComfy. You can pass any of the flags supported upstream by ComfyUI directly into your machine at launch time. The complete, authoritative list of arguments lives in the official ComfyUI source code: comfy/cli_args.py.

This FAQ walks through the three places where startup arguments can be set, persisted, and reused on RunComfy.

⚠️ Tip: Invalid or incompatible startup arguments can prevent ComfyUI from booting, break specific features, or noticeably slow down execution — double-check every flag against cli_args.py before saving or deploying. ⚠️ Tip: We highly recommend that you do not add listen 127.0.0.1 yourself.

1. Set arguments at launch via Advanced Settings

When you spin up a ComfyUI machine, expand the Advanced Settings panel on the launch screen. Any flags you enter here (for example --lowvram, --use-pytorch-cross-attention, --preview-method auto) are passed directly to the ComfyUI process at boot. Refer to cli_args.py for the full set of supported flags and their accepted values.

2. Arguments are preserved with Cloud Save

When you trigger a Cloud Save, the startup arguments used in the current session are stored alongside the rest of your workspace state. The next time you launch from that save:

• The previous arguments are pre-filled automatically, so you don't have to re-enter them.
• You can clear them entirely and launch with default settings.
• You can overwrite them with a completely new set of arguments.

3. Arguments travel with workflow Deployments

When you deploy a workflow, the startup arguments captured for that specific workflow version are applied automatically on every deployment run. This guarantees that the runtime environment matches the one the workflow was authored and tested against.

As with Cloud Save, you remain in full control — the arguments can be reviewed, edited, cleared, or replaced before any run.

Available ComfyUI Startup Arguments

Below is the complete list of ComfyUI startup arguments currently supported by RunComfy. Arguments that require or accept a value are shown with their expected value format.

Note: These arguments come directly from ComfyUI. Some options are mutually exclusive, experimental, hardware-specific, or may affect performance and stability. Only change them if you know what they do.

Network & Server

--listen [IP]
--port PORT
--tls-keyfile TLS_KEYFILE
--tls-certfile TLS_CERTFILE
--enable-cors-header [ORIGIN]
--max-upload-size MAX_UPLOAD_SIZE

Directories & Paths

--base-directory BASE_DIRECTORY
--extra-model-paths-config PATH [PATH ...]
--output-directory OUTPUT_DIRECTORY
--temp-directory TEMP_DIRECTORY
--input-directory INPUT_DIRECTORY
--user-directory USER_DIRECTORY
--models-directory MODELS_DIRECTORY

Launch Behavior

--auto-launch
--disable-auto-launch

GPU & Device

--cuda-device DEVICE_ID
--default-device DEFAULT_DEVICE_ID
--cuda-malloc
--disable-cuda-malloc
--directml [DIRECTML_DEVICE]
--oneapi-device-selector SELECTOR_STRING
--supports-fp8-compute
--enable-triton-backend
--disable-triton-backend

Model Precision

--force-fp32
--force-fp16

--fp32-unet
--fp64-unet
--bf16-unet
--fp16-unet
--fp8_e4m3fn-unet
--fp8_e5m2-unet
--fp8_e8m0fnu-unet

--fp16-vae
--fp32-vae
--bf16-vae
--cpu-vae

--fp8_e4m3fn-text-enc
--fp8_e5m2-text-enc
--fp16-text-enc
--fp32-text-enc
--bf16-text-enc

--fp16-intermediates
--force-channels-last

Preview

--preview-method [none,auto,latent2rgb,taesd]
--preview-size PREVIEW_SIZE

Caching

--cache-ram [GB ...]
--cache-classic
--cache-lru CACHE_LRU
--cache-none
--high-ram

Attention

--use-split-cross-attention
--use-quad-cross-attention
--use-pytorch-cross-attention
--use-sage-attention
--use-flash-attention
--disable-xformers
--force-upcast-attention
--dont-upcast-attention

ComfyUI Manager

--enable-manager
--disable-manager-ui
--enable-manager-legacy-ui

VRAM & Memory

--gpu-only
--highvram
--lowvram
--novram
--cpu

--reserve-vram RESERVE_VRAM
--vram-headroom VRAM_HEADROOM
--disable-nvml-pressure

--async-offload [NUM_STREAMS]
--disable-async-offload

--disable-dynamic-vram
--enable-dynamic-vram

--fast-disk
--force-non-blocking
--disable-smart-memory
--disable-pinned-memory
--mmap-torch-files
--disable-mmap

Performance & Execution

--deterministic
--fast [FAST ...]
--debug-hang
--default-hashing-function {md5,sha1,sha256,sha512}

Valid optimizations for --fast currently include:

fp16_accumulation
fp8_matrix_mult
cublas_ops
autotune

Running --fast without specifying an optimization enables all available --fast optimizations.

Custom Nodes & API Nodes

--disable-all-custom-nodes
--whitelist-custom-nodes WHITELIST_CUSTOM_NODES [WHITELIST_CUSTOM_NODES ...]
--disable-api-nodes

Server & User Behavior

--dont-print-server
--quick-test-for-ci
--windows-standalone-build
--disable-metadata
--multi-user

Logging

--verbose [LEVEL FILE ...]
--log-stdout

Frontend

--front-end-version FRONT_END_VERSION
--front-end-root FRONT_END_ROOT
--enable-compress-response-body

ComfyUI API & Database

--comfy-api-base COMFY_API_BASE
--database-url DATABASE_URL

Assets

--enable-assets
--enable-asset-hashing

Feature Flags

--feature-flag KEY[=VALUE]
--list-feature-flags

• -feature-flag can be specified multiple times. For example:

--feature-flag show_signin_button=true

or:

--feature-flag show_signin_button

Help

-h
--help

For detailed descriptions, accepted values, defaults, and the latest upstream changes, always refer to the official ComfyUI cli_args.py.

References

• ComfyUI CLI argument source of truth: github.com/Comfy-Org/ComfyUI/blob/master/comfy/cli_args.py
• RunComfy: www.runcomfy.com