Local LoRA Gallery Creator (V5)

A local, single file HTML gallery for browsing, organizing, and managing your Stable Diffusion LoRA collection. No web server, no installation beyond Python, just run the script and open the HTML in your browser.

Built for people with large LoRA collections who want to explore, compare, and select styles outside of their SD UI (ComfyUI, A1111, etc).

This was built primarily for personal use. Updates are not guaranteed, but issues and PRs are welcome.

Requirements

• Python 3.8 or later, no external packages required (stdlib only)
• A folder of LoRA files (.safetensors, .pt, or .ckpt)
• Optional: CivitAI .civitai.full.info or .civit.info metadata files alongside your LoRAs, for names, triggers, tags, creator info, likes, and downloads

Quick Start

1. Copy LocalLoraGalleryV5.py and Rebuild Gallery V5.bat into your LoRA root folder
2. Double click Rebuild Gallery V5.bat
3. Choose option 1 (Smart rebuild) and press Enter
4. When prompted, press Y to open the gallery in your browser

The gallery is fully self contained in lora_gallery.html. Open it directly in any browser, no server needed.

How it works

The script works like a compiler:

LoRA folders + images + metadata
        |
Python scan and parse
        |
Static HTML gallery
        |
Browser based UI (search, notes, favorites)

You only need to rerun the script when files on disk change. Notes, favorites, and presets do not require regeneration.

Bat Menu Options

[1] Smart rebuild            Rescan folders, reuse cached prompt metadata, use this daily
[2] Skip-prompts rebuild     Rescan folders only, skip all image metadata, fastest
[3] Full rebuild             Rescan and re-read ALL PNG prompts from scratch, thorough
[4] Clear cache + rebuild    Delete metadata cache, then smart rebuild, after updates
[5] Open gallery             Open existing gallery instantly, no scan, quick browse
[6] Update CivitAI metadata  Re-fetch stats and triggers from CivitAI API, refresh data
[7] Fetch CivitAI images     Download top sample images per LoRA, get previews
[8] Run deletions            Execute run_deletions.py to move marked items, cleanup

Folder Structure

The script scans recursively from where it lives. It works with any nesting, flat, one level deep, or multiple levels. The recommended layout is:

loras/
  LocalLoraGalleryV5.py
  Rebuild Gallery V5.bat
  Anima/
    SomeCharacter/
      SomeCharacter.safetensors
      SomeCharacter.civitai.full.info   (CivitAI metadata, optional)
      preview.png                        (preview image, optional)
  Pony/
    AnotherLora/
      ...
  SDXL/
    ...

The first folder level (Anima, Pony, SDXL, etc) becomes the category shown in the sidebar.

CivitAI Metadata (.info files)

LoRAs downloaded via CivitAI Helper or similar tools often include .civitai.full.info files. The gallery reads these automatically to display:

• Trigger words
• Tags
• Base model (Pony, SDXL, Illustrious, etc)
• Creator name
• Download count and likes
• CivitAI model link

Without a .info file, the gallery still shows the LoRA, just without the above metadata.

Updating CivitAI Metadata (Option 6)

Refreshes stats, trigger words, and tags from the CivitAI API for any LoRA that already has a .info file.

• Interactive folder picker, choose one subfolder (e.g. just "Pony") or all
• Detects changes: [NEW VERSION], [UPDATED], [+triggers], [-triggers]
• Saves a _civitai_update_report.txt with links to changed models on civitai.red
• API delay of 0.1s between requests

LoRAs without a .info file cannot be updated since there is no ID to look up.

Fetching Sample Images (Option 7)

Downloads the top 5 most reacted images from CivitAI for each LoRA and saves them into the LoRA's folder.

• Interactive folder picker, target one subfolder or all
• Images saved as _civitai_top_1.jpg, _civitai_top_2.jpg, etc.
• Prompt metadata saved as _civitai_top_1.jpg.params (A1111 format), visible in the gallery lightbox
• Skips images already downloaded, safe to re-run
• NSFW images included by default
• Automatically rebuilds the gallery after downloading

Gallery Features

Search and Filter

• Search bar with AND term search across folder names, filenames, triggers, tags, creator, base model
• "+Notes" also searches your personal notes
• "Img dupes" toggles duplicate prompt images on or off
• Filters: Has Triggers, Favs, Hide NSFW, NSFW threshold, Dupes only, Unused only
• Base model dropdown to filter to one base model
• Sort by Name, Newest, Folder, Triggers, Most Used, NSFW Level, Largest, Fav: Recent, Most Liked, Most Downloaded

Sidebar

• Favorites, with a live count, click to filter to favorites only
• Recently Used, last 8 LoRAs you interacted with, clickable to jump
• Collections, custom labels you add to cards (e.g. character, style, wip), click to filter
• Categories, auto built from your top level folders

Cards

• Click the card title to add to the prompt queue
• CivitAI link on every card with metadata (opens civitai.red)
• Star icon to favorite, [18+] to manually tag NSFW, "+ Label" to add collections
• Copy filename, copy <lora:stem:1> tag, add to prompt queue
• Trigger pills, click any to copy, "Copy all" copies all at once
• Notes textarea, saved in your browser, searchable with the +Notes checkbox
• Image strip with variable height slider (drag in toolbar)

Lightbox

• Click any image to open full screen
• Shows positive and negative prompts, Steps, CFG, Sampler, Seed, Size, Model
• Works for local PNGs with embedded metadata and for CivitAI downloaded images (via .params sidecar)
• Buttons: Copy prompt, Copy negative, Send to sandbox, Copy all params
• Keyboard shortcuts: left and right arrows to navigate, Esc to close

Prompt Sandbox (right panel)

• LoRA Queue, add LoRAs, set weights (0.1 to 2.0), click "+ Tags & Triggers" to build your prompt
• Presets, save and load named prompt and queue combinations
• History, last 20 prompts you copied
• Quick Inserts, model specific tag buttons (Universal, Illustrious, Anima, Pony)
• Wildcards, browse and insert from wildcard .txt files

Duplicate Management

Duplicate LoRA folders (same model, different folders)

The gallery detects LoRAs with the same CivitAI model ID or similar name and shows a yellow DUPE badge. Use the "Dupes only" filter to find them. Use the "Delete folder" button to mark one for deletion, then run deletions.

Duplicate images (same prompt, different files)

Within each folder, if two images have identical generation prompts, the older one is:

• Hidden by default (use the "Img dupes" checkbox to show them)
• Auto marked for deletion with a yellow outline
• Shown in a strip note, for example "N duplicate images hidden, marked for deletion"

To keep a specific duplicate, click its close button, which sets a "keep" flag so it will not be re-marked.

Deletion Workflow

The gallery never permanently deletes anything, it moves files to _DELETED/.

1. Mark items in the gallery, click the close button on images or "Delete folder" on LoRA rows; duplicate images are auto marked
2. Check the count, the "Deletions (N)" button in the toolbar shows the total marked
3. Download the script, click "Deletions (N)" to save run_deletions.py to your browser's downloads folder, then move it to your loras folder
4. Run option 8 in the bat to execute run_deletions.py, which moves everything to _DELETED/
5. Rebuild (option 1) to remove them from the gallery
6. Permanently delete _DELETED/ only once you have confirmed everything looks right

Compact Mode

Click "Compact" in the toolbar to switch to a thumbnail grid view. This hides trigger and tag rows, notes, and action buttons. The thumbnail size slider still works in compact mode. The CivitAI link remains visible.

Wildcards

Place .txt files in a wildcards/ subfolder next to the script. Each file becomes a browsable category in the right panel. You can also drag and drop .txt files directly into the wildcard drop zone in the gallery (session only).

Extra Prompt Pack

Create a prompt_pack.json file in the same folder as the script to add custom quick insert buttons. The structure mirrors the built in packs:

{
  "universal": {
    "quality": ["my custom tag", "another tag"],
    "negative": ["unwanted thing"]
  }
}

Data and Privacy

All your personal data (favorites, notes, usage counts, labels, queue, presets) is stored in your browser's localStorage. It never leaves your machine. The gallery HTML can be shared with others and they will start with a clean slate.

The CivitAI metadata update and image fetch features make outbound HTTP requests to civitai.com/api/v1/. No other network requests are made.

Tips

• First run on a large collection: use option 3 (Full rebuild) once, then option 1 for all subsequent rebuilds
• Added new LoRAs: option 1 is sufficient, it only re-reads metadata for new or changed files
• Prompts missing in lightbox: run option 4 (Clear cache + rebuild), previous cache entries may be stale
• Downloaded CivitAI images not showing prompts: run option 4, the sidecar check was added in V5 and old cache entries may say "no metadata"
• Slow fetch: target a specific subfolder when using options 6 and 7 instead of updating all folders at once
• civitai.red: all links in the gallery open on civitai.red, a mirror. CivitAI API calls still go to civitai.com

Recommended Workflow

This tool pairs well with the Firefox Civitai Model Downloader by Anton:

https://addons.mozilla.org/en-US/firefox/addon/civit-model-downloader/

That extension downloads the LoRA file, preview images, and full Civitai metadata. Drop those folders directly into your Stable Diffusion LoRA directory and this script will scan them as is to build the gallery automatically.