Image MetaHub

Local-first AI image organizer and generative media library manager for ComfyUI, Automatic1111, InvokeAI, Forge, and more.

Image MetaHub indexes large folders of AI-generated media without moving or uploading your files. Search and organize generations by prompt, checkpoint, LoRA, sampler, seed, workflow metadata, tags, ratings, or visual similarity, then send compatible workflows straight back into ComfyUI.

Join the community on Discord — feature requests, bug reports, metadata parsing help, and early builds.

Image MetaHub main UI

What is Image MetaHub?

Image MetaHub is a desktop app for organizing and browsing large local libraries of AI-generated images, video, audio, and other generative media. It is built for people who have outgrown a normal output folder or basic ComfyUI gallery and need a searchable library that still understands how each generation was made.

It scans your existing folders in place, extracts generation metadata from ComfyUI, Automatic1111, InvokeAI, Forge, and other tools, caches results for fast reuse, and lets you combine metadata search, filters, tags, ratings, collections, lineage, and local visual similarity search. Your files stay on your machine.

Organize ComfyUI generations at scale

Image MetaHub works as a standalone ComfyUI output organizer for large generation folders. It reads embedded prompt and workflow metadata, lets you search by checkpoint, LoRA, sampler, scheduler, seed, dimensions, node types, tags, ratings, and more, and can send compatible generations back into ComfyUI as working workflows.

You do not need to import your library into a proprietary cloud or reorganize your output folders first. Image MetaHub indexes the files where they already live and can keep watching those folders while ComfyUI generates new media.

Manage large AI-generation libraries

Image MetaHub is more than an output gallery. It is designed for long-term retrieval and curation across large local collections, with metadata search, visual similarity, tags and ratings, stacks, collections, lineage, compare tools, deduplication helpers, and support for multiple generation ecosystems.

The goal is to preserve the production history around each asset: what prompt, model, LoRAs, workflow, and source image created it, how related generations connect to each other, and how to continue working from it later.

Highlights

Local-first browsing with no mandatory account, no cloud sync, and no outbound telemetry
Fast indexing and thumbnail caching for large libraries
Find Similar — Local Visual Search for discovering related images even when they have no prompt or generation metadata
Metadata parsing for Automatic1111, ComfyUI, InvokeAI, SD.Next, Forge, SwarmUI, Fooocus, Draw Things, Midjourney/Niji, Firefly, DreamStudio, DALL-E, and more
Support for PNG, JPG, JPEG, WEBP, AVIF, GIF, MP4, WEBM, MKV, MOV, and AVI
Faceted sidebar filters with explicit include/exclude actions for checkpoints, LoRAs, samplers, schedulers, ratings, generation modes, media types, ComfyUI node types.
Unified Explore surface for drilling into Models, Clusters, and Collections, with the drill-in kept as a combinable filter chip
Embedded ComfyUI Workspace with a live ComfyUI browser, library thumbnails, workflow metadata, and quick generation actions
Live step-by-step generation previews in the queue for ComfyUI jobs
Image lineage detection for img2img, inpaint, and outpaint, including source-image recovery when possible
Multi-window image viewer with move, resize, docking/collapsible details, and fast cross-reference workflows
Built-in image editor with adjustments, crop, transform, resize, enhance, annotations, and metadata-preserving export
Prompt clustering, TF-IDF auto-tags, manual tag management, and deduplication helpers
Startup verification modes for reopening saved libraries from cache or validating them against disk
Automatic1111 and ComfyUI integrations with queueing, progress tracking, and optional launcher shortcuts
Opt-in Civitai lookups for model and LoRA hashes, with results cached locally
Analytics Explorer and verified metrics support for images generated with the MetaHub Save Node

Free vs Pro

The repository is MPL 2.0 and the core app remains open-source. Some workflow-heavy features are unlocked through the desktop app's offline Pro license or 7-day trial.

Core app includes:

Local indexing, metadata parsing, search, sort, and filtering
Tags, favorites, safe mode, and shadow metadata editing
Auto-watch for generation folders
Image lineage display and multi-window viewer workflows
Explore workspace with auto-tags and clustering under free-tier limits
Built-in image editor
Deduplication helpers and stack browsing

Pro currently unlocks:

Automatic1111 generation and parameter copy workflows
ComfyUI generation, embedded workspace, workflow-native editing, and progress tracking
Compare View with 2-4 image layouts and metadata diff tools
Analytics Explorer
Batch export
Bulk tagging
In-app file management (copy/move between indexed folders)
Unlimited clustering scale
Unlimited Local Visual Search indexing

Getting Started

Download the latest desktop release from GitHub Releases.
Install and launch Image MetaHub.
Add one or more folders that contain your generated images or videos.
Wait for the first indexing pass to finish.
Use search, sidebar facets, tags, and advanced filters to explore the library.

Windows installer and portable build

Windows releases provide a standard Setup executable and a separate Portable executable. The portable build keeps its settings, browser data, metadata cache and thumbnails in an ImageMetaHubData folder beside the executable, without using the Image MetaHub profile in AppData.

To update the portable build, close Image MetaHub, replace only the Portable executable, and keep the existing ImageMetaHubData folder beside it.

macOS unsigned builds

Current GitHub release builds are not signed with an Apple Developer ID yet. If macOS blocks the app after you download and move it to Applications, remove the quarantine flag from Terminal:

xattr -dr com.apple.quarantine "/Applications/Image MetaHub.app"

This is a temporary workaround for unsigned builds until macOS signing and notarization are available.

Browsing and filters

Browsing and Curation

Image MetaHub is built around fast local curation:

Search + facets: combine free-text search with include/exclude facets for checkpoints, LoRAs, samplers, schedulers, ComfyUI node types, tags, favorites, ratings, generation modes, media types, and advanced ranges
Stacking: group identical prompts in the main library view for faster browsing
Sort + Group By: sort order and grouping live in the persistent grid footer, with grouping by date, name, generation session, checkpoint model, or cluster, plus Jump To navigation
Manual tags + ratings: keep a persistent manual tag catalog, switch included tags between Any and All, and curate with 1-5 ratings
Metadata recovery: reparse selected images without running a full folder refresh or clearing cache — a single reparse patches only the affected cache chunk, so it stays fast on large libraries
Image editor: adjust, crop, rotate, flip, resize, sharpen, blur, annotate, and redact from the viewer, then Save As a metadata-preserving copy or overwrite eligible originals on desktop
Startup verification: choose whether saved folders reopen from cache, reconcile in the background, or verify strictly before startup completes
Shadow metadata: edit metadata non-destructively and keep the original payload available for inspection or revert
Viewer workflows: in the desktop app, open each image in a separate native window above Image MetaHub while the ComfyUI Workspace stays alive; minimize and restore viewers from the footer, or select the legacy in-app viewer in Settings
Auto-watch: keep output folders in sync while A1111 or ComfyUI is generating

Find Similar — Local Visual Search

Select any image and find visually related results across your library, including files with no prompt or generation metadata. Processing runs locally using an optional on-device model.

Local Visual Search is off by default. Enabling it does not download anything: the model download is a separate, explicit action, and downloaded files are verified before use. Indexing can be paused and resumed, works offline after the model is installed, and uses optional WebGPU acceleration with an automatic WASM fallback.

Find Similar is the primary supported workflow in v0.19. Experimental text-to-image queries are also available, but their results are approximate. They do not replace Image MetaHub's deterministic search across prompts, metadata, models, LoRAs, and workflows.

Explore

Explore is the single discovery workspace that replaced the separate Model View, Smart Library, and Collections screens. It presents your library across three dimensions — Models, Clusters, and Collections — as browsable cards.

Scope drill-in: opening a card scopes the Library grid to it and shows the scope as a dedicated chip in Active Filters, so it stays combinable with every other filter instead of trapping you in a separate screen
Prompt clustering: a background worker groups related images by prompt similarity
TF-IDF auto-tags: generate useful tags from prompts, models, LoRAs, and workflow metadata
Stack browsing: open a stack, paginate inside it, and keep navigation context in the image viewer
Deduplication helper: rank likely keep/archive candidates and estimate space savings
Back button: return from a drill-in scope straight to the matching Explore dimension
Free-tier limits: the open-source app includes Explore workflows, while Pro removes clustering scale limits

Library-wide operations (automation rules, auto-tagging the library, analytics) live in the Library Tools menu in the library toolbar.

Classic Mode

If you prefer the previous navigation, Settings → Appearance → Classic Mode restores the old Model View / Smart Library / Collections / Node View labels as shortcuts that deep-link into Explore.

ComfyUI Nodes Filter

The standalone Node View has been retired in favor of a multi-select ComfyUI Nodes filter in the sidebar's Generation Parameters, for libraries with embedded ComfyUI workflows:

Search exact node-type names across the current scope
Multi-select node filters with OR matching
See per-node result counts before applying a filter
Combine node filters with every other filter and the active scope, without leaving the library grid

Metadata Support

Image MetaHub reads metadata from:

Stable Diffusion / Automatic1111 images
ComfyUI workflows and prompt graphs
InvokeAI
SD.Next
Forge
Fooocus
SwarmUI
Draw Things
Midjourney / Niji Journey
Adobe Firefly
DreamStudio
DALL-E
Other tools that embed generation parameters into PNG/JPEG/WebP/AVIF metadata or sidecar payloads

Supported media types:

Images: PNG, JPG, JPEG, WEBP, AVIF, GIF
Video: MP4, WEBM, MKV, MOV, AVI
Audio: MP3, WAV, FLAC, OGG, OGA, M4A, AAC, OPUS, AIFF, AIF, WMA

For video and audio metadata, Image MetaHub uses container metadata plus ffprobe when available to extract duration, codec, frame count, resolution, sample rate, channels, and bit rate. Audio files can be indexed even when a codec is not playable by Chromium/Electron on the current system.

AVIF metadata

Image MetaHub reads ComfyUI prompt and workflow documents from standard AVIF XMP items, including files produced by the proposed ComfyUI AVIF saver and frontend loader. It also reads the older EXIF convention used by existing ComfyUI AVIF files.

New Image MetaHub AVIF exports are intentionally compact: the full prompt and workflow graphs stay in their established ComfyUI XMP fields rather than being copied again. The Image MetaHub namespace stores app-specific fields (tags, notes, attribution, analytics, lineage, source generator) alongside the extracted parameter snapshot (model, seed, steps, cfg, sampler, scheduler, negative prompt) the Save Node records — those structured fields are kept because re-deriving them from arbitrary custom-node graphs is unreliable. When a legacy nested prompt conflicts with a standalone XMP prompt, the standalone document wins and the viewer shows a metadata-conflict badge.

Civitai links (opt-in)

Model and LoRA hashes in the Image Modal are clickable and open the matching Civitai page. The lookup only happens when you click — a single request to Civitai's public API, cached locally so each hash is looked up at most once. It works with A1111, Forge, SD.Next, and Fooocus images, plus ComfyUI images saved with the MetaHub Save Node, and can be disabled entirely under Settings → Privacy. Indexing and browsing stay fully offline either way.

MetaHub Save Node

For ComfyUI, the best experience comes from the companion ImageMetaHub Save Node on the ComfyUI Registry.

With the Save Node, Image MetaHub can ingest:

Full workflow and prompt payloads
Tags and notes saved by the workflow
GPU and timing analytics
Explicit lineage metadata for derived images

For older ComfyUI images without the node, Image MetaHub still attempts best-effort parsing from embedded workflow data.

Image details and metadata

Image Lineage and Viewer

The image viewer is no longer just a single modal. In the current app it supports:

Multiple open image windows at the same time
Move, resize, minimize, maximize, and focus management
Docked or collapsed detail panels
Lineage display for transformations like img2img, inpaint, and outpaint, including generation-type and denoise context when available
Source-image recovery from explicit references or inferred metadata when possible
Derived-image previews to navigate transformation chains

Automatic1111 Integration (Pro)

With Pro enabled, Image MetaHub can talk directly to a running Automatic1111 instance.

Main workflows:

Copy to A1111: format metadata into A1111's three-line parameter block for the blue-arrow import flow
Generate with A1111: send normalized metadata directly to the API for quick regeneration
Model and LoRA selection: browse available models/LoRAs and override prompt parameters before generation
Queue-aware progress: generations feed the shared queue and progress surfaces

Basic setup:

Start A1111 with --api.
If needed, allow the app origin with --cors-allow-origins=http://localhost:5173.
Configure the server URL in Image MetaHub settings.

ComfyUI Integration (Pro)

With Pro enabled, Image MetaHub can generate through ComfyUI using either the original embedded workflow or a safe metadata rebuild.

The ComfyUI Workspace adds a full working area around a running local ComfyUI instance: an embedded browser, selected-image context, library thumbnails, directory scoping, workflow/raw metadata tabs, copy/generate actions, and direct routes from the grid, table, toolbar, and image viewer. In the browser build, the same entry points open ComfyUI externally instead of embedding it.

Current flow:

Open an image with compatible metadata.
Click Generate with ComfyUI.
Choose Original workflow or Simple rebuild.
Adjust prompt, negative prompt, seed, steps, CFG, dimensions, model overrides, LoRAs, and source image policy when relevant.
Optionally use the visual workflow editor or advanced JSON editor.
Queue the workflow to ComfyUI and follow progress in real time over WebSocket.

What exists today:

Workflow-native mode for executable embedded prompt graphs
Simple rebuild mode for metadata-only images
Visual workflow inspector/editor with pan/zoom and per-node field editing
Embedded ComfyUI Workspace for staying inside Image MetaHub while browsing source images and working in ComfyUI
Model-family aware overrides for checkpoints, UNETs, VAEs, CLIP loaders, and LoRAs when supported
Transform-aware source image policies for img2img/inpaint-style workflows
Shared queue with retry, cancel, cleanup actions, and optional detection of ComfyUI jobs started outside Image MetaHub
Live generation preview in the queue, updating step-by-step during generation for jobs started from MetaHub and from the embedded ComfyUI UI, with a resizable preview box that remembers its height
Run current workflow straight from the queue, queueing whatever workflow is loaded in the embedded ComfyUI workspace
Metadata-rich outputs when used with the MetaHub Save Node and Timer node

Setup:

Run ComfyUI locally, usually on http://127.0.0.1:8188.
Install the MetaHub Save Node.
Configure the ComfyUI URL in settings.
Optionally save a local launch command so the desktop header can start or reopen ComfyUI for you.
Test the connection from the app and start generating.

Compare View (Pro)

Image MetaHub currently supports comparing up to 4 images with:

Side-by-side mode with optional synchronized zoom/pan
Side Strip and 2x2 Grid layouts for 3-4 image sets
Slider, hover, flicker, difference map, loupe, and edge modes for two-image comparisons
Metadata comparison in standard or diff view, with all panels expanding, collapsing, and scrolling in sync
Quick swap and keyboard shortcuts

Compare panel

Analytics (Pro)

The Analytics Explorer summarizes library usage and generation performance:

Overview, Resources, Time, Performance, and Curation views
Scope switching between the current filtered results and the full library
Cohort comparisons for generators, models, LoRAs, samplers, GPU devices, ratings, and more
One-click promotion of analytics insights into live filters
Average speed, VRAM, and generation time for MetaHub Save Node images
Performance charts grouped over time or by GPU

Analytics dashboard

Development

This repository contains the desktop app source code.

Stack:

React 18 + TypeScript
Electron
Zustand
Vite
Tailwind CSS
Vitest

Common commands:

npm install
npm run dev
npm run dev:app
npm run build
npm run electron-dist
npm run test
npm run lint

CLI helpers:

npm run cli:parse -- path/to/file.png --pretty --raw
npm run cli:index -- path/to/folder --out index.jsonl --recursive

For release work, see RELEASE-GUIDE.md and RELEASE-AUTOMATION.md.

Troubleshooting

On macOS, if audio or video playback crashes the Electron Helper, quit Image MetaHub first, then launch it from Terminal with the opt-in media safe mode:

IMH_MEDIA_SAFE_MODE=1 /Applications/Image\ MetaHub.app/Contents/MacOS/Image\ MetaHub

To confirm the flag was applied, process-events.log should include mediaSafeModeEnabled: true in the app-startup entry. This disables hardware acceleration plus accelerated video decode/compositing switches for that launch only. The lower-level IMH_DISABLE_GPU=1 flag is also available when you want to test GPU acceleration separately.

Privacy

Image MetaHub is designed to stay local:

Your files, cache, tags, and metadata stay on your machine.
There is no required account system.
Pro licenses are validated offline.
Network activity is limited to things that explicitly need it, such as:
- auto-update checks
- local A1111 / ComfyUI APIs
- Civitai hash lookups you explicitly click (can be disabled in Settings → Privacy)
- links you choose to open

Credits

Image MetaHub is built and maintained by Lucas (LuqP2) with community feedback and contributions.