CivitAI Integration Overview

The CivitAI integration spans multiple backend modules and provides several capabilities:

Capabilities

1. Catalog cross-reference -- map CivitAI model/version IDs to local catalog entries, checking download status
2. Add models from CivitAI -- fetch a CivitAI model version's metadata and add it to the local catalog without downloading the file
3. Promote LoRAs -- move a LoRA entry from models.json to loras.json (promoting it to a style LoRA)
4. Image generation data -- proxy CivitAI's tRPC API to fetch generation parameters for any image, enriched with local catalog status
5. Dependency detection -- parse <lora:Name:weight> tags from prompts and embeddings from negative prompts, then resolve them via CivitAI hash lookup
6. Workflow type detection -- heuristically determine if an image was generated with t2i, i2i, i2v, or t2v
7. Gallery downloads -- download CivitAI images and videos for LoRA galleries

Backend Modules

Module	Responsibilities
civitai_api.py	Generation data proxy, catalog add, promote-to-lora, dependency resolution, workflow type detection
loras_api.py	LoRA lookup, gallery download/serve, CivitAI tag sync
models_api.py	CivitAI map endpoint, metadata fetch from CivitAI
gallery_db.py	SQLite index for downloaded gallery images

Authentication

CivitAI API calls use the CIVITAI_API_KEY environment variable. The key is sent as a Bearer token in the Authorization header. Without this key:

• Model version fetches may fail for restricted content
• Image generation data may be unavailable for some images
• Hash-based model resolution still works for public models

The key is configured either as an environment variable on the pod or edited at runtime from the Settings page.