Configuration

There are three layers of configuration, all readable from shared:

1. Command-line flags — parsed once at boot in modules/cmd_args.py. Available as shared.cmd_opts. Read-only.
2. User options — declared in modules/shared_options.py, persisted to config.json. Available as shared.opts. Edited via the Settings tab or POST /sdapi/v1/options.
3. UI defaults — slider/dropdown initial values stored in ui-config.json by modules/ui_loadsave.py.

This page lists the most-used flags and the categories of options. The exhaustive list lives in the source.

Command-line flags

Roughly 130 flags. Defined with parser.add_argument(...) in modules/cmd_args.py. Common ones:

Network / surface

| Flag | Effect | | ----------------------------------------------------------------- | ----------------------------------------------------- | -------------------------- | | --listen | Bind to 0.0.0.0 | | --port N | Listen on port N (default 7860, 7861 for --nowebui) | | --api | Enable the FastAPI routes alongside the UI | | --nowebui | Run as headless API server | | --share | Open a Gradio public tunnel | | --ngrok TOKEN | Tunnel via ngrok | | --subpath PATH | Host under a sub-path | | --cors-allow-origins=URLs, --cors-allow-origins-regex=PATTERN | CORS allow list | | --gradio-auth USER:PASS, --gradio-auth-path FILE | UI HTTPBasic | | --api-auth USER:PASS, --api-auth-file FILE | API HTTPBasic | | --api-server-stop | Register /sdapi/v1/server-stop | | --api-log | Log every API request | | --enable-insecure-extension-access | Allow extension installs over the network | | --disable-all-extensions all | extra | Disable extensions at boot | | --disable-extra-extensions | Disable user extensions, keep built-ins | | --tls-keyfile FILE, --tls-certfile FILE | TLS termination | | --gradio-allowed-path PATH | Whitelist additional paths Gradio can serve |

Hardware / memory

| Flag | Effect | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------ | --------- | ------------------------------------ | | --xformers | Enable xFormers cross-attention | | --force-enable-xformers | Enable even if compatibility check fails | | --xformers-flash-attention | xFormers + Flash Attention (SD2.x only) | | --opt-split-attention, --opt-split-attention-v1, --opt-split-attention-invokeai, --opt-sub-quad-attention, --opt-sdp-attention, --opt-sdp-no-mem-attention, --disable-opt-split-attention | Pick attention backend | | --medvram, --medvram-sdxl, --lowvram | VRAM optimisations | | --lowram | Load weights to VRAM rather than RAM | | --no-half | UNet in fp32 | | --no-half-vae | VAE in fp32 | | --upcast-sampling | Upcast just the sampler math | | --precision full | half | autocast | Default precision (default autocast) | | --use-cpu MODULE [MODULE ...] | Force specific submodules to CPU (all to force everything) | | --use-ipex | Intel XPU | | --skip-torch-cuda-test | Don't fail if CUDA isn't available | | --disable-nan-check | Don't reject NaN-producing models |

Paths

Flag	Effect
--data-dir PATH	Override the data directory (config, outputs, …)
--models-dir PATH	Override the models directory
--ckpt FILE	Use a specific checkpoint
--ckpt-dir DIR, --vae-dir DIR, --embeddings-dir DIR, --hypernetwork-dir DIR	Override individual model dirs
--ui-config-file PATH, --ui-settings-file PATH	Override config files

Debug / dev

Flag	Effect
--loglevel LEVEL	Log level
--log-startup	Verbose boot
--gradio-debug	Gradio's verbose mode
--dump-sysinfo	Write a sanitised sysinfo JSON and exit
--reinstall-torch, --reinstall-xformers	Force reinstall on next boot
--skip-prepare-environment, --skip-install	Skip env prep
--update-all-extensions, --update-check	Update extensions / check for updates
--allow-code	Enable the custom-code script
--test-server	Use a stub model for tests
--do-not-download-clip	Don't auto-download CLIP

For the complete list, run python launch.py --help or read modules/cmd_args.py.

User options (opts)

Declared in modules/shared_options.py (~50 KB, ~400 settings). Each call to options_templates.update(options_section(...)) adds a section visible in the Settings tab.

The sections, in order:

Section	Examples
Saving images	filename pattern, jpeg/avif quality, save grids
Paths for saving	per-tab output dirs
Saving to a directory	always-save, paste-info
Upscaling	ESRGAN tile size, RealESRGAN models
Face restoration	enable, weight
System	memory monitor, GIT executable, profiler
Profiling	torch profiler options (added v1.8)
API	api_enable_requests, api_forbid_local_requests, api_useragent
Training	TI / hypernet defaults
Stable Diffusion	active checkpoint, VAE, clip-skip, hires-fix defaults
SDXL	refiner ratio, etc.
Compatibility	legacy emphasis, old hires-fix
VAE	active VAE, fp32
img2img	inpainting masking, color correction
Optimizations	token-merging, hypertile, AND-composable defaults
Sampler parameters	UniPC variant, eta, schedule type defaults
Hidden	settings the user can show/hide from the Settings UI
Postprocessing	always-apply postprocessors after generation
Live previews	preview mode (Approx vs TAESD), every-N steps
Defaults	UI defaults written to ui-config.json
User interface	quicksettings list, hires-fix layout, theme
Infotext	what to write into PNG metadata
Settings	localisation, keyboard shortcuts
Gallery	columns, filters
Extra Networks	card size, sort, refresh

Each option declares an OptionInfo with default, label, component class, and optional onchange and comment_after (for tooltips). Extensions add their own via script_callbacks.on_ui_settings.

The on-disk format of config.json is a flat JSON dict from option key to value:

{
  "samples_save": true,
  "outdir_samples": "",
  "sd_model_checkpoint": "v1-5-pruned-emaonly.safetensors [6ce0161689]",
  "...": "..."
}

UI defaults

ui-config.json records "what the user has clicked on for each control". When the user changes a slider or dropdown, the new value is written to this file (if opts.save_ui_settings_to_file is on). On the next page load, modules/ui_loadsave.py restores those values.

This file is per-user and machine-local. It is not versioned with the repo.

Other config files in the data dir

File	Purpose
config.json	opts
ui-config.json	UI control defaults
cache.json	Model hashes, extension commit info, sampler caches
styles.csv	Saved prompt styles
params.txt	Last-used generation parameters
tmp/	Gradio temp uploads
log/	Generation log

The data dir is --data-dir (defaults to repo root).

Environment variables

A handful of env vars are honoured outside of cmd_args.py:

Var	Effect
COMMANDLINE_ARGS	Read by webui-user.sh/.bat into the launch flags
TORCH_COMMAND	Override the pip command used to install torch
XFORMERS_PACKAGE	Override the xFormers pip command
REQS_FILE	Override the requirements file path
python_cmd, venv_dir	Used by webui.sh/.bat for venv setup
SD_WEBUI_LOG_LEVEL	Default log level
SD_WEBUI_RESTARTING	Set to 1 during in-process UI reload to suppress auto-launch
IIB_LIST	Used by some extensions for the image browser; not core