Models and LoRAs

Checkpoint Models & LoRA Library

Managing hundreds of Stable Diffusion checkpoint models and fine-tuned LoRAs is a challenge in generative AI workflows. Berry AI Studio provides built-in cataloging, hash resolution, and trigger word management.

---

1. Checkpoint Model Catalog (ModelManagerModal.vue)

Berry automatically tracks every checkpoint model encountered across your library.

Automatic Model Discovery:

• When indexing files, Berry parses checkpoint names and model hashes from embedded PNGInfo/EXIF metadata.
• Open Tools > Model Manager & Cache to view a catalog of all models, their short hashes, full SHA256 checksums, and total image counts.

Resolving Hashes with AUTOMATIC1111 cache.json:

• Checkpoint models often appear as 8-character hashes (e.g. 31e35c80).
• If you have an existing AUTOMATIC1111 installation:
  1. Click "Import Model Cache" in the Model Manager.
  2. Select your WebUI cache.json file (typically located in <webui_root>/cache.json).
  3. Berry imports the mapping into its local model_cache table, instantly resolving cryptic hashes across your entire library into friendly model titles.

Civitai SHA256 Hash Resolution:

• For models without local names, you can click the Civitai lookup button in the Inspector to query Civitai's public model database via model hash.

---

2. LoRA Trigger Words Library (LoraManagerModal.vue)

Low-Rank Adaptations (LoRAs) require specific activation or trigger words in generation prompts to produce intended character details, art styles, or clothing.

┌────────────────────────────────────────────────────────────────────────┐
│ LoRA Trigger Words Library & Manager                               [✕] │
├────────────────────────────────────────────────────────────────────────┤
│ [🔍 Search LoRAs... ]                   [+ Add LoRA] [📁 Scan Directory]│
├────────────────────────────────────────────────────────────────────────┤
│ LoRA Name               │ Default Wt │ Trigger Words          │ Actions │
├─────────────────────────┼────────────┼────────────────────────┼─────────┤
│ CyberpunkCityStyle      │ 0.80       │ cyberpunk, neon signs, │ [Copy]  │
│                         │            │ futuristic alleys      │ [Inject]│
│ GenshinRaidenShogun     │ 0.85       │ raiden shogun, purple  │ [Copy]  │
│                         │            │ braid, glowing katana  │ [Inject]│
│ StudioGhibliVintage     │ 0.70       │ ghibli style, vintage  │ [Copy]  │
│                         │            │ watercolor, cel shaded │ [Inject]│
└────────────────────────────────────────────────────────────────────────┘

Key LoRA Features:

1. Detected LoRAs in Inspector:
   • When viewing an artwork, the Property Inspector automatically sniffs <lora:name:weight> prompt syntax and ComfyUI LoraLoader nodes.
   • It lists detected LoRAs, their applied weights, and recognized trigger words.
2. One-Click Prompt Injection:
   • Click "+ In Prompt" to copy the formatted <lora:name:0.8> string directly to your clipboard.
   • Click any trigger word chip to copy it for immediate use in your prompt editor.
3. Importing Civitai Sidecars (.civitai.info):
   • If you download LoRAs with Civitai helper sidecars (.civitai.info or .json), Berry extracts model hashes, base model architectures (SD 1.5, SDXL, Pony, Flux), trained trigger words, and preview images automatically.
4. Local Directory Scanning:
   • Point Berry to your local LoRA folder (models/Lora/).
   • Berry scans all .safetensors files and companion preview images, building a comprehensive searchable offline reference library.