Logo Vincent
Back to all posts

StableDiffusionWebUI: Startup Arguments

AI Art
StableDiffusionWebUI: Startup Arguments

Preface

sdwebui has many startup arguments. This article analyzes them one by one.

Startup Arguments

The argument parsing is located in modules/cmd_args.py

-f: Allow root user to start

—update-all-extensions: Update all extensions on startup

—skip-python-version-check: Skip Python version check

—skip-torch-cuda-test: Skip CUDA check. If you don’t have a GPU locally, you can skip this first. CUDA is a computing platform by NVIDIA.

—reinstall-xformers: Reinstall xformers. xformers is an attention mechanism that can optimize the image generation process.

—reinstall-torch: Reinstall torch. Torch is a machine learning library.

—update-check: Check for updates on startup

—test-server: Start in test environment

—log-startup: Print detailed log information on startup

—skip-prepare-environment: Skip all environment preparation steps, such as checking various environments

—skip-install: Skip all dependency installation steps

—dump-sysinfo: Export system information to local

—loglevel: Set log level

—do-not-download-clip: Do not download the default CLIP model

—data-dir: Set user data path

—config: Set configuration file path

—ckpt: Checkpoint model file path. If set, this model will be loaded automatically.

—ckpt-dir: Checkpoint model directory

—vae-dir: VAE directory

—gfpgan-dir: GFPGAN directory. Deprecated, use —gfpgan-models-path instead.

—gfpgan-model: GFPGAN model path. Deprecated, use —gfpgan-models-path instead.

—no-half: Do not convert models to 16-bit floats

—no-half-vae: Do not convert VAE to 16-bit floats

—no-progressbar-hiding: Do not hide the Gradio progress bar

—max-batch-count: Maximum batch count

—embeddings-dir: Embeddings directory

—textual-inversion-templates-dir: Textual inversion templates directory

—hypernetwork-dir: Hypernetwork directory

—localizations-dir: Localizations directory

—allow-code: Allow custom scripts to execute in the WebUI

—medvram: Sacrifice some image generation speed to reduce VRAM usage

—medvram-sdxl: Enable medvram option only for SDXL

—lowvram: Sacrifice significant image generation speed to reduce VRAM usage

—lowram: Sacrifice significant image generation speed to reduce RAM usage

—always-batch-cond-uncond: Does nothing?

—unload-gfpgan: Does nothing?

—precision: Set precision

—upcast-sampling: Upcast sampling mode, improves image generation efficiency and reduces VRAM usage

—share: Share the project externally

—ngrok: Ngrok authentication configuration for sharing

—ngrok-region: Same as above

—ngrok-options: Same as above

—enable-insecure-extension-access: Enable the Extensions tab

—codeformer-models-path: CodeFormer directory

—gfpgan-models-path: GFPGAN directory

—esrgan-models-path: ESRGAN directory

—bsrgan-models-path: BSRGAN directory

—realesrgan-models-path: Real-ESRGAN directory

—dat-models-path: DAT directory

—clip-models-path: CLIP directory

—xformers: Enable xformers

—force-enable-xformers: Force enable xformers

—xformers-flash-attention: Enable xformers with flash attention

—deepdanbooru: Does nothing?

—opt-split-attention: Use doggettx to optimize the image generation process

—opt-sub-quad-attention: Optimization related

—sub-quad-q-chunk-size: Optimization related

—sub-quad-kv-chunk-size: Optimization related

—sub-quad-chunk-threshold: Optimization related

—opt-split-attention-invokeai: Optimization related

—opt-split-attention-v1: Optimization related

—opt-sdp-attention: Optimization related

—opt-sdp-no-mem-attention: Optimization related

—disable-opt-split-attention: Optimization related

—disable-nan-check: Disable NaN check

—use-cpu: Use CPU mode

—use-ipex: Use Intel XPU mode

—disable-model-loading-ram-optimization: Disable model loading RAM optimization

—listen: By default, only requests from localhost are accepted. Adding this argument accepts all network requests.

—port: Set port number

—show-negative-prompt: Does nothing?

—ui-config-file: UI configuration file, defaults to ui-config.json

—hide-ui-dir-config: Hide UI directory configuration

—freeze-settings: Disable modification of all global settings

—freeze-settings-in-sections: Disable modification of a specific settings section

—freeze-specific-settings: Disable modification of specific settings

—ui-settings-file: UI settings file, defaults to config.json

—gradio-debug: Start Gradio in debug mode

—gradio-auth: Enable Gradio authentication

—gradio-auth-path: Gradio auth file path

—gradio-img2img-tool: Does nothing?

—gradio-inpaint-tool: Does nothing?

—gradio-allowed-path: Add files to Gradio’s accessible paths

—opt-channelslast: Change SD memory to channels last mode

—styles-file: Styles file, supports multiple files

—autolaunch: Automatically open in browser after startup

—theme: Dark or light mode

—use-textbox-seed: Use textbox to input seed

—disable-console-progressbars: Do not output progress in console

—enable-console-prompts: Does nothing?

—vae-path: VAE path

—disable-safe-unpickle: Disable the check for malicious code in models

—api: Also start in API mode

—api-auth: Enable API authentication

—api-log: Enable API log output

—nowebui: Do not start WebUI, only start API mode

—ui-debug-mode: Do not load models, for quickly starting the UI

—device-id: Specify CUDA device ID

—administrator: Start with admin privileges

—cors-allow-origins: Enable CORS

—cors-allow-origins-regex: Enable CORS with regex matching

—tls-keyfile: TLS key file

—tls-certfile: TLS certificate file

—disable-tls-verify: Disable TLS verification

—server-name: Server name

—gradio-queue: Does nothing?

—no-gradio-queue: Disable WebSocket requests, use HTTP polling instead

—skip-version-check: Skip version check, including torch and xformers

—no-hashing: Skip hash check to improve model loading speed

—no-download-sd-model: Do not download the default model

—subpath: Subpath

—add-stop-route: Does nothing?

—api-server-stop: Allow stopping and restarting the service via API

—timeout-keep-alive: Set timeout, defaults to 30s

—disable-all-extensions: Disable all extensions

—disable-extra-extensions: Disable all extensions except built-in ones

—skip-load-model-at-start: Do not load models on startup

—unix-filenames-sanitization: Allow Unix filenames

—filenames-max-length: Maximum filename length

—no-prompt-history: Do not record prompt history

SDWebUI

|—URL: https://sdwebui.ai/?from=vq

|—Authentic Stable Diffusion WebUI V1.10.1~

|—Top-tier 4090 GPU, premium members get exclusive single 4090 for image generation~

|—Supports Mac client, Windows client, and Web online usage~

|—Innovative instant model transfer, upload local models instantly~

|—Supports txt2img, img2img, extras, and image info

|—Supports CLIP interrogation and DeepBooru interrogation

|—Supports HD upscaling (Hires.fix), with 21 built-in upscaling models

|—Supports detail restoration (ADetailer), with 11 built-in restoration models

|—Supports ControlNet, with 44 built-in preprocessors and 68 built-in models

|—Supports ultra-HD upscaling (Tiled Diffusion, Demo Fusion, Tiled VAE)

|—Supports old photo restoration, with 21 built-in upscaling models, plus GFPGAN and CodeFormer restoration models

|—Built-in 2 common Styles, 11 popular Checkpoint models, 7 popular LoRA models, 14 common embeddings

|—Built-in oldsix prompt plugin

|—Built-in 23-lesson beginner tutorial

© 2026 Vincent. All rights reserved.