---

name: launch-and-config description: "Launch Stable Diffusion WebUI safely in UI, API, and API-only modes, choose configuration flags, and diagnose startup failures." disable-model-invocation: true metadata: disco-role: operating license: AGPL 3.0

Launch and Config

Use this sub-skill when a task asks to start WebUI, enable the REST API, run API-only mode, set model/data paths, configure auth/TLS/CORS/subpath, reduce VRAM use, avoid launcher side effects, or diagnose Python/torch/xformers/checkpoint startup failures.

Fast Start

1. Pick the server mode:
   • UI only: python launch.py --ckpt <checkpoint.safetensors>
   • UI plus API: python launch.py --api --api-auth user:pass --ckpt <checkpoint.safetensors>
   • API only: python launch.py --nowebui --api-auth user:pass --ckpt <checkpoint.safetensors>
2. Choose path flags before launch: --data-dir, --models-dir, --ckpt, --ckpt-dir, --vae-dir, --embeddings-dir, --ui-settings-file, and --ui-config-file.
3. Add network exposure flags deliberately: --listen or --server-name, --port, --gradio-auth, --api-auth, --tls-keyfile, --tls-certfile, --subpath, and only narrow --cors-allow-origins values.
4. Add hardware/performance flags from evidence: --medvram, --medvram-sdxl, --lowvram, --lowram, --xformers, --opt-sdp-attention, --upcast-sampling, --no-half, --no-half-vae, --use-cpu, or --device-id.
5. Diagnose startup with troubleshooting.md before bypassing checks; --skip-python-version-check, --skip-torch-cuda-test, --skip-install, and --skip-prepare-environment hide symptoms and can leave dependencies unresolved.

Bundled References

• launch-and-installation.md: launch entry points, side effects, mode examples, wrapper behavior, and safe command patterns.
• configuration.md: flag catalog by purpose, config files, environment variables, security posture, path strategy, and reverse-proxy setup.
• troubleshooting.md: failure signals and fixes for Python, torch/CUDA/xformers, missing checkpoints, downloads, auth/CORS/TLS/subpath, config freeze, and path mistakes.
• extract_cli_flags.py: self-contained AST helper that extracts argparse flags from a WebUI modules/cmd_args.py file without importing WebUI.

Example helper usage when auditing a checkout:

python sub-skills/launch-and-config/scripts/extract_cli_flags.py --source modules/cmd_args.py --format markdown
python sub-skills/launch-and-config/scripts/extract_cli_flags.py --source modules/cmd_args.py --format json

Safe Defaults

• Do not expose a server with --listen, --server-name 0.0.0.0, --share, or --ngrok unless auth and network controls are explicit.
• Pair UI auth and API auth separately: --gradio-auth protects Gradio UI; --api-auth protects /sdapi/v1/* routes.
• Treat --allow-code, --enable-insecure-extension-access, --disable-safe-unpickle, and --api-server-stop as high-risk; use only for trusted local environments.
• Use --no-download-sd-model when startup must not fetch a fallback checkpoint; provide --ckpt or --ckpt-dir instead.
• Use --disable-all-extensions or --disable-extra-extensions while debugging launch failures from user extensions.

Boundaries

• Route REST payload construction, endpoint semantics, and API smoke requests to ../api-automation/SKILL.md.
• Route checkpoint, VAE, Lora, embedding, upscaler, hash, and asset layout details to ../assets-and-models/SKILL.md.
• Route extension authoring, callback hooks, script templates, and extension install behavior beyond launch gating to ../extension-scripting/SKILL.md.
• Route textual inversion, hypernetwork training, extras, face restoration, and postprocessing workflows to ../training-and-postprocessing/SKILL.md.