Skip to content

Feature Extra Networks

eduardoabreu81 edited this page Jul 24, 2026 · 1 revision

Extra Networks

Detection, autocomplete, and metadata handling for LoRA, LyCORIS, Textual Inversion, and embeddings.

  • Detection — recognized chips are highlighted with distinct colors per network type, with an existence check against your actual model folders and a metadata popup showing preview image and trained keywords
  • Autocomplete — LoRA/LyCORIS triggered by <, embeddings by <e:, with thumbnail previews
  • ⭐ Correct alias insertion — the identifier inserted matches whatever Forge Neo actually expects (respecting the lora_preferred_name setting: filename vs. alias), so tokens never "blink" between two representations after insertion
  • ⭐ Trigger word injection — on LoRA selection, trigger words are pulled from optional CivitAI Browser Neo metadata (if installed) or Prompt Studio's own SHA-256 cache, and inserted at a configurable position in the prompt. Model sidecar files (.json, .civitai.info) are read-only inputs — never written to.

Backend sharing

Prompt Studio owns one model metadata cache (SHA-256, base model, trigger words, CivitAI IDs) shared between the tag autocompletion engine and Quality Presets, and reuses Forge Neo's own extra-network file catalog for paths/aliases/hashes instead of scanning LoRA directories on every request. See Quality Presets for how the same metadata backend feeds checkpoint family detection.

Clone this wiki locally