L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Données et recherche · SKILL.md

assets-and-models

assets-and-models

Manage Stable Diffusion WebUI checkpoints, VAEs, embeddings, hypernetworks, Lora/extra networks, upscalers, model paths, refresh endpoints, metadata, and asset troubleshooting.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: assets-and-models description: "Manage Stable Diffusion WebUI checkpoints, VAEs, embeddings, hypernetworks, Lora/extra networks, upscalers, model paths, refresh endpoints, metadata, and asset troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: AGPL 3.0

Assets and Models

Use this sub-skill when the task is about placing, discovering, refreshing, validating, or diagnosing WebUI model assets rather than generating images or changing launch/security policy.

Route Here For

  • Checkpoint, VAE, embedding, hypernetwork, Lora/LyCORIS, and upscaler directory questions.
  • Custom --data-dir, --models-dir, --ckpt-dir, --vae-dir, --embeddings-dir, --hypernetwork-dir, --lora-dir, or upscaler model-path flag issues.
  • API/UI refresh and list questions for checkpoints, VAE, Lora, embeddings, hypernetworks, and upscalers.
  • Safetensors metadata, checkpoint hashes, Lora hashes, missing asset cards, incompatible extra networks, or upscaler fallback diagnostics.

Fast Decisions

  • Prefer .safetensors for checkpoints, VAEs, and Lora when possible; .ckpt/.pt are supported in several places but may rely on pickle loading.
  • Default model root is models/ under --data-dir; --models-dir replaces that root, while --ckpt-dir, --vae-dir, --lora-dir, and upscaler path flags add or override specific searches.
  • Refresh after placing files: use the relevant UI refresh button or API endpoint from model-and-asset-layout.md.
  • Lora prompt syntax is <lora:name:weight> with optional text-encoder, UNet, and dynamic-dimension overrides; see extra-networks-and-lora.md.
  • When diagnosing without loading models, run validate_asset_layout.py against the intended data/models directories first.

Runtime References

  • Model and asset layout lists default directories, launch flags, recognized suffixes, API endpoints, hashes, metadata, VAE resolution, and upscaler directories.
  • Extra networks and Lora covers Lora/LyCORIS discovery, prompt forms, aliases, hashes, metadata, and incompatibility signals.
  • Troubleshooting gives concrete triage paths for missing checkpoints, VAE discovery, hash mismatch, unsafe formats, Lora visibility, and upscaler fallback.
  • Layout validator script safely checks directory layout and recognized file suffixes with no WebUI imports and no model loading.

Boundaries

  • For txt2img/img2img payload fields, generation API schemas, or switching checkpoints during generation, route to api-automation.
  • For launch-flag selection, networking, API exposure, authentication, or extension security, route to launch-and-config.
  • For creating/training embeddings, hypernetworks, or checkpoints, route to training-and-postprocessing.
  • For implementing or debugging extension code internals, route to extension-scripting.
Skills enregistrées