视觉相似度搜索需先单独显式下载模型，否则无法使用；macOS 构建未签名，移入 Applications 后可能需执行 xattr 命令解除隔离；Pro 功能（生成、模型管理、ComfyUI 工作区等）需授权，试用仅 7 天。
Image-MetaHub
LuqP2活跃维护本地优先的 AI 图像整理与生成媒体库管理工具，支持 ComfyUI、A1111、InvokeAI 等，可按提示词、模型、LoRA、元数据或视觉相似度检索海量输出目录。
Local-first AI image organizer and generative media library manager for ComfyUI, A1111, InvokeAI & more. Search huge output folders by prompt, model, LoRA, metadata or visual similarity.
v0.22.1发布于 2026-10-22 12:01 UTC335star26fork同步于 2026-11-10 12:01 UTC
npm install
npm run dev:app 硬件要求按 README 说明
README 未说明具体显存或 GPU 型号要求；视觉相似度索引支持可选 GPU 加速，也可在本地 CPU 上运行，音视频播放依赖系统编解码器。
本地优先的 AI 图像库管理工具，按提示词、模型、LoRA 或视觉相似度检索生成结果
本地优先的 AI 图像整理与生成媒体库管理工具，支持 ComfyUI、A1111、InvokeAI 等，可按提示词、模型、LoRA、元数据或视觉相似度检索海量输出目录。
核心能力
- 按提示词、模型、LoRA、采样器等元数据搜索并组合过滤
- 用本地视觉相似度检索，无需生成元数据也能找图
- 解析 ComfyUI、A1111、InvokeAI 等多家工具的生成元数据
- Prompt Library 保存、收藏并跨工具导入导出提示词
- 内置对比、图像编辑、模型管理与 ComfyUI 工作区（Pro）
容易踩的坑
内容由模型根据 README 生成，可能不完整或不准确，以官方文档为准。看原文 README
- v0.22.1Image MetaHub v0.22.1最新正式版2026-10-22 · 2026-10-22 12:01 UTC
Image MetaHub v0.22.1
[0.22.1] - [2026-10-22]
Added
- OpenEXR / OpenColorIO Support: Added support for browsing, organizing, and previewing
.exrfiles throughout the Library, image viewer, detached viewer, and comparison tools. Includes color-managed previews with ACES 1.3, ACES 2.0, and custom OCIO configurations; global and per-folder color profiles; non-destructive exposure adjustments; multipart EXR inspection; render passes and individual channels; and high-resolution region previews when zooming. The optional component can be installed from Settings → General → Optional components and works offline after installation.
Improved
- ComfyUI Workspace Layout: Moved navigation and connection controls into the main application header, removed redundant workspace footer content, and added direct access to the generation queue with a pending-job indicator. The streamlined layout provides more vertical space for ComfyUI workflows.
- Prompt Library Preview Images: Added the ability to attach, replace, and remove preview images directly from saved prompt and block details. Choose images from the Library using a searchable picker with filename, prompt, and model matching.
- Prompt Library Inline Editing: Titles, categories, tags, and notes can now be edited directly from the details panel without opening the full editor. Added searchable category and tag selectors with inline creation, streamlined preview controls, and a dedicated content editing action.
- macOS Update Notifications: Updated the macOS update flow to open GitHub Releases directly, allowing users to download and install the latest DMG manually. Update notifications now clearly indicate when manual installation is required.
Fixed
- Library Scroll Position: Fixed unwanted scroll resets when new files are detected by folder monitoring. Background additions no longer interrupt browsing or return the Library to the beginning. Navigation and filter changes, including new visual searches, continue to reset the position appropriately.
- Library Multi-Selection Preview: Fixed the preview sidebar not following the most recently Ctrl-clicked or Cmd-clicked image when selecting multiple items.
- ComfyUI ClownsharKSampler Metadata: Fixed positive prompt, negative prompt, and model extraction from RES4LYF
ClownsharKSamplerandClownsharKSampler_Betaworkflows. Improved tracing through linked text inputs and Krea2 conditioning nodes while preventing unrelated or disabled samplers from overriding the correct metadata. - Prompt Preview Picker: Fixed missing thumbnails when choosing preview images and ensured Safe Mode filtering and sensitive-image blurring are respected.
Downloads
Choose the appropriate installer for your operating system:
Windows
- Installer:
ImageMetaHub-Setup-0.22.1.exe - Format: NSIS installer with desktop and start menu shortcuts
- Size: ~85MB
macOS
- Intel Macs:
ImageMetaHub-0.22.1.dmg - Apple Silicon:
ImageMetaHub-0.22.1-arm64.dmg - Format: DMG packages with proper entitlements
- Requirements: macOS 10.15+
Linux
- Universal:
ImageMetaHub-0.22.1.AppImage - Format: Portable AppImage (no installation required)
- Dependencies: None (fully self-contained)
System Requirements
- OS: Windows 10+, macOS 10.15+, Ubuntu 18.04+ (or equivalent)
- RAM: 4GB minimum, 8GB recommended
- Storage: 100MB for application + space for your image collections
Documentation
Known Issues
- Safari, Firefox, and Brave browsers don't support the File System Access API on macOS
- Use Chrome, Vivaldi, Edge, or the Desktop App for full functionality
Feedback
Found a bug or have a feature re
- OpenEXR / OpenColorIO Support: Added support for browsing, organizing, and previewing
- exr-runtime-v1.0.0EXR / OpenColorIO runtime 1.0.02026-10-21 · 2026-10-21 13:02 UTC
查看更新说明
Optional EXR / OpenColorIO runtime for Image MetaHub.
Install explicitly from Settings > EXR / OpenColorIO in a compatible app build. Processing runs locally and works offline after installation.
Includes Windows x64, Linux x64, macOS x64 and macOS arm64 packages, dependency notices, and SHA-256 integrity policies. All four packages passed native numerical tests and isolated installation/offline-preview/removal smoke tests.
This is a component release, not an Image MetaHub application update.
- v0.22.0Image MetaHub v0.22.02026-10-17 · 2026-10-17 20:23 UTC
查看更新说明
Image MetaHub v0.22.0
[0.22.0] - [2026-10-17]
Added
- Reusable Prompt Blocks and Templates (Pro): Expanded the Prompt Library with reusable text blocks, prompt composition, and fillable templates. Combine positive and negative prompts, arrange blocks with drag-and-drop, insert multiple saved items, and preview the final result. Block instances preserve their content independently of later changes to the originals.
- Prompt Library Organization: Added tags, favorites, notes, advanced filters, bulk actions, exact and similar duplicate detection, and JSON /
styles.csvimport and export. - ComfyUI Workspace Context Tools (Pro): Added Image, Prompts, and Models tabs to the Workspace sidebar. Browse saved prompts and blocks, compose new prompts, access model trigger words, and send text directly to compatible fields in selected ComfyUI nodes.
- ComfyUI Prompt Drag-and-Drop (Pro): Drag saved prompts, blocks, or composed text directly into ComfyUI node fields. Drop to append text, or hold Shift while dropping to choose between Add and Replace.
- Find Images from Prompts: Find matching Library images directly from saved prompts and blocks, including text fragments embedded within longer generation prompts.
Improved
- Prompt Library Experience: Redesigned browsing and editing with list and grid views, a dedicated details panel, inline tag editing, searchable multi-item selection, reusable block suggestions, and a live composition preview. Unsaved drafts are preserved while navigating between library sections.
- ComfyUI Workflow Integration: Save selected node text directly as prompts or blocks, append or replace existing node text, and preserve workflow history with undo/redo support. Compatible actions provide immediate feedback and protect against overwriting changed node values.
- Ayu Themes and Branding: Added Ayu Light and made the Ayu theme family the default for system appearance. Refreshed application logos and icons, with improved contrast across light themes and workspace controls.
- Library Navigation and Controls: Stabilized the main navigation layout, introduced text-only section tabs, consolidated Library toolbar actions, and made active filters more compact.
- Model Manager Visibility: Added the ability to hide models from the main catalog and Workspace sidebar. Hidden models are also excluded from automatic update checks and notification counts.
Changed
- Source Code Licensing: Updated the project's current source license from MPL 2.0 to PolyForm Perimeter 1.0.1. Previously distributed versions retain their original license grants.
Fixed
- Large Library Processing and Recovery: Metadata enrichment now saves progress incrementally, reducing lost work after interruptions. The main application window can recover from renderer crashes or out-of-memory failures and resume unfinished processing from saved checkpoints.
- ComfyUI MiniMax H3 Metadata: Added support for extracting prompts, model information, and generation parameters from MiniMax H3 workflows, including multiframe configurations and linked conditioning nodes.
- ComfyUI Disconnected Samplers: Fixed incorrect or missing metadata when disconnected sampler nodes appeared after the active generation sampler in a workflow.
- Prompt Preservation: Fixed handling of negative-only prompts and edited shadow metadata when saving or composing prompts, preserving available source and generation information.
Downloads
Choose the appropriate installer for your operating system:
Windows
- Installer:
ImageMetaHub-Setup-0.22.0.exe - Format: NSIS installer with desktop and start menu shortcuts
- Size: ~85MB
macOS
- Intel Macs:
ImageMetaHub-0.22.0.dmg - Apple Silicon:
ImageMetaHub-0.22.0-arm64.dmg - Format: DMG packages with proper entitlements
- Requirements: macOS 10.15+
Linux
- Universal: `ImageMetaHub-0.
- v0.21.0Image MetaHub v0.21.02026-10-09 · 2026-10-09 00:44 UTC
查看更新说明
Image MetaHub v0.21.0
[0.21.0] - [2026-10-08]
Added
- Model Manager (Pro): Introducing a dedicated workspace for managing local LoRAs, checkpoints, and other AI models. Organize your collection with covers, favorites, tags, notes, and custom metadata; identify models through Civitai or link them to Hugging Face; track new releases and file changes; inspect models in a detachable window; find associated images in your Library; and manage storage with duplicate detection and file removal.
- Masonry Grid Layout: Added an optional Library layout that displays thumbnails at their original aspect ratios, with virtualized scrolling, grouping, keyboard navigation, and multi-selection.
- Extended Pro Trial: Users whose original Pro trial has expired can activate an additional three days to explore the latest features. The one-time offer is optional and starts only when activated.
Improved
- Detached Viewer: Added an option to reuse a single viewer window when opening different images, instead of creating multiple windows. Available under Settings → Viewer → Behavior.
Fixed
- Detached Viewer Reliability: Fixed issues with failed or interrupted window launches, image selection synchronization, and reopening viewers.
- macOS Large Library Monitoring: Fixed file descriptor exhaustion when monitoring large libraries, which could prevent detached viewer windows from opening.
Downloads
Choose the appropriate installer for your operating system:
Windows
- Installer:
ImageMetaHub-Setup-0.21.0.exe - Format: NSIS installer with desktop and start menu shortcuts
- Size: ~85MB
macOS
- Intel Macs:
ImageMetaHub-0.21.0.dmg - Apple Silicon:
ImageMetaHub-0.21.0-arm64.dmg - Format: DMG packages with proper entitlements
- Requirements: macOS 10.15+
Linux
- Universal:
ImageMetaHub-0.21.0.AppImage - Format: Portable AppImage (no installation required)
- Dependencies: None (fully self-contained)
System Requirements
- OS: Windows 10+, macOS 10.15+, Ubuntu 18.04+ (or equivalent)
- RAM: 4GB minimum, 8GB recommended
- Storage: 100MB for application + space for your image collections
Documentation
Known Issues
- Safari, Firefox, and Brave browsers don't support the File System Access API on macOS
- Use Chrome, Vivaldi, Edge, or the Desktop App for full functionality
Feedback
Found a bug or have a feature request? Open an issue!
Released on 2026-10-09
- v0.20.1Image MetaHub v0.20.12026-10-04 · 2026-10-04 05:36 UTC
查看更新说明
Image MetaHub v0.20.1
[0.20.1] - [2026-10-04]
Improved
- Detached Viewer Navigation: Added visible previous and next controls to detached viewer windows. Navigation now keeps the Library's active image, card highlight, and preview sidebar synchronized with the image currently being viewed.
- Smart Library Cluster Restore: Saved clusters are now detected earlier during startup, with a loading state shown while they are being restored instead of briefly presenting an empty cluster view.
Fixed
- Detached Viewer Lifecycle: Opening another image while a detached viewer is already open or still loading is now handled reliably. Delayed viewer windows no longer steal focus, closing the active viewer returns cleanly to the Library, and the Library remains on the last image viewed.
- Detached Viewer Favorites: The favorite heart now immediately reflects the actual saved favorite state instead of appearing inverted until the viewer is refreshed.
- Smart Library Cluster Persistence: Saved clusters now restore correctly when new images have been added since the cache was created, while still rejecting stale clusters when previously indexed image metadata has changed. Cluster restoration also completes correctly for empty libraries and libraries without clusterable prompts.
Downloads
Choose the appropriate installer for your operating system:
Windows
- Installer:
ImageMetaHub-Setup-0.20.1.exe - Format: NSIS installer with desktop and start menu shortcuts
- Size: ~85MB
macOS
- Intel Macs:
ImageMetaHub-0.20.1.dmg - Apple Silicon:
ImageMetaHub-0.20.1-arm64.dmg - Format: DMG packages with proper entitlements
- Requirements: macOS 10.15+
Linux
- Universal:
ImageMetaHub-0.20.1.AppImage - Format: Portable AppImage (no installation required)
- Dependencies: None (fully self-contained)
System Requirements
- OS: Windows 10+, macOS 10.15+, Ubuntu 18.04+ (or equivalent)
- RAM: 4GB minimum, 8GB recommended
- Storage: 100MB for application + space for your image collections
Documentation
Known Issues
- Safari, Firefox, and Brave browsers don't support the File System Access API on macOS
- Use Chrome, Vivaldi, Edge, or the Desktop App for full functionality
Feedback
Found a bug or have a feature request? Open an issue!
Released on 2026-10-04
用户点评
登录用户的真实使用记录，可按版本和硬件筛选共 0 条
还没有人点评。第一个说出真实体验的人最有价值。
Image MetaHub
Find, organize, and reuse your AI generations — all on your own machine.
Image MetaHub turns your output folders into a searchable desktop library. Find an image by its prompt, model, LoRA, or visual similarity; save prompts for later; and continue working in ComfyUI or Automatic1111. Your files stay where they are, with no uploads or account required.
Download the app · Website · Get Pro · Discord
Get started
- Download the build for your system from GitHub Releases.
- Open Image MetaHub and add your generation folders.
- Let the first scan finish, then browse, search, and filter your library.
There is no need to move your files or change your folder structure. Enable folder watching to keep the library updated as new generations arrive.
Windows: choose Setup for a standard installation or Portable to keep the app's settings and data beside the executable. When updating Portable, replace only the executable and keep its ImageMetaHubData folder.
macOS: current builds are unsigned. Update notifications open GitHub Releases, where you can download the latest DMG and install it manually. If macOS blocks the app after you move it to Applications, run this in Terminal:
xattr -dr com.apple.quarantine "/Applications/Image MetaHub.app"
Find the generation you had in mind
Search prompts and metadata, then narrow the results by checkpoint, LoRA, sampler, tags, rating, folder, media type, or ComfyUI node. Include what you want and exclude what you don't; active filters stay visible above the results.
Browse in grid, Masonry, or table view. Use favorites, ratings, manual tags, and Auto-Tag to keep useful work easy to find. Explore groups images into models, prompt clusters, and collections, while stacks and duplicate helpers make large batches easier to curate. Themes include Ayu, Ayu Light, and additional choices in Settings.
Find Similar searches for visually related images, including files with no generation metadata. It runs locally after a separate, explicit model download, with pausable indexing and optional GPU acceleration. Text-to-image visual queries are also available as an experimental option.
Inspect an image and recover how it was made
Open an image to read its prompt, model, LoRAs, seed, and other available generation settings. Use a separate viewer window, keep several references open, or reuse a single window as you browse.
For related generations, lineage can show source images and transformation chains such as img2img, inpaint, and outpaint. Provenance summarizes available source evidence and offers an optional SHA-256 fingerprint. Shadow metadata lets you edit information while keeping the original available to restore.
Keep your best prompts ready to reuse
Save a prompt from any image into Prompt Library, then search, favorite, copy, or return to its source. Find images takes you back to library images matching a saved prompt or block. Import and export through JSON or styles.csv to move your prompts between tools.
With Pro, turn useful fragments into reusable blocks, compose positive and negative prompts, and create templates with fillable fields. Saved compositions keep their own block snapshots, so later edits to a block won't silently change your earlier work.
Manage your local AI models · Pro
The Models workspace organizes checkpoints, LoRAs, diffusion models, VAEs, text encoders, and other local model files with covers, tags, favorites, notes, and trigger words.
Identify versions through Civitai, link files to Hugging Face, and follow new releases or file changes in What's new. Both services have independent, optional monitoring controls. Checks retrieve information without downloading or replacing your model weights.
See which library images use a model, inspect it in a detached window, or open Storage to review file sizes, locations, and verified duplicate copies. File removal lets you review the selected paths before moving them to the system Trash.
Continue creating in ComfyUI or Automatic1111 · Pro
Open ComfyUI Workspace to work with a running local ComfyUI instance inside the app, alongside your library images, saved prompts, blocks, and models. Reuse prompt text in compatible node fields or save selected node text for later.
Generate from a compatible original workflow or rebuild a simpler workflow from image metadata. Inspect nodes, adjust parameters, and follow queued jobs with live previews. Automatic1111 integration also lets you copy generation parameters or send a generation directly to its API.
Connect your running generator in Settings. For richer ComfyUI outputs, the companion MetaHub Save Node preserves workflow metadata, tags, notes, and available performance metrics. Generation requires the appropriate models and nodes in your generator.
Arrange references on Canvas · Pro
Open Canvas to create a saved board, or select library assets and choose Add to Board. Drag references from its Library panel, arrange them with text, frames and arrows, and double-click an asset to open its viewer. Select two to four images to compare them with the existing Compare View.
Boards save automatically and keep references to your original files. Export or import an .imhboard.json document to back up a layout; media files remain in your library. Existing boards stay viewable and exportable after your trial ends.
Compare, edit, and understand your results · Pro
Compare up to four images, with synchronized zoom and metadata differences. Two-image modes include a slider, flicker, loupe, difference map, and edge comparison.
The built-in Image Editor provides adjustments, crop, rotation, resizing, sharpening, and annotations, with metadata-preserving exports.
Analytics Explorer shows which models, LoRAs, samplers, and generators you use, across the full library or current results. Images with MetaHub Save Node telemetry can also contribute generation-time, speed, GPU, and VRAM metrics.
Images, video, audio, and 3D in one library
Image MetaHub reads generation metadata from ComfyUI, Automatic1111, InvokeAI, Forge, SD.Next, SwarmUI, Fooocus, Draw Things, Midjourney/Niji, Firefly, DreamStudio, DALL-E, and other compatible exports. Available details depend on what the source tool saved.
| Media | Supported formats |
|---|---|
| Images | PNG, JPG/JPEG, WEBP, AVIF, GIF; EXR with the optional desktop component |
| Video | MP4, WEBM, MKV, MOV, AVI |
| Audio | MP3, WAV, FLAC, OGG/OGA, M4A, AAC, OPUS, AIFF/AIF, WMA |
| 3D | GLB, GLTF, OBJ, FBX, STL |
Browse and play media, or inspect 3D assets in the interactive viewer and export them. Audio/video playback depends on the codecs available on your system. Files without generation metadata can still be browsed and organized.
For renders, EXR / OpenColorIO adds color-managed previews, render passes, individual channels, and non-destructive exposure. Install the optional component from Settings → General → Optional components; processing then works offline. EXR originals are preserved, and EXR editing/export and deep images are outside this first version.
Free and Pro
The core library is free to use. Pro adds generation, model management, and more tools for larger workflows. A 7-day Pro trial is available on eligible desktop installations.
| Feature | Free | Pro |
|---|---|---|
| Local browsing, metadata search, filters, tags, ratings, and favorites | Included | Included |
| Folder watching, viewer windows, lineage, and shadow metadata | Included | Included |
| Basic Prompt Library, import/export, and exact prompt duplicate detection | Included | Included |
| Local Visual Search indexing | 2,000 newest images | Unlimited |
| Prompt clustering | Limited scale | Unlimited |
| Prompt block creation, composition, and template editing | — | Included |
| Model Manager and model update tracking | — | Included |
| ComfyUI Workspace and Automatic1111 integration | — | Included |
| Image Editor, Compare, and Analytics | — | Included |
| Batch export, bulk tagging, and in-app file copy/move | — | Included |
If Pro or a trial expires, existing prompt blocks and templates remain readable, fillable, copyable, and exportable.
Privacy
Your library, prompts, tags, and metadata stay on your machine. Browsing and indexing work locally, with no cloud sync or outbound telemetry. Local Visual Search also works offline after its optional model is downloaded.
Network access is used for updates, services you connect to, and optional actions such as Civitai/Hugging Face lookups, model monitoring, and downloads. Pro licenses are verified locally.
Help and development
Join Discord for questions, metadata parsing help, and feature requests, or report a bug through GitHub Issues. See the changelog for release history.
To run the desktop app from source:
npm install
npm run dev:app
Developer references: Architecture · CLI · Prompt Library · ComfyUI context tools · Release guide
Built and maintained by Lucas (LuqP2) with community feedback and contributions.
License
Image MetaHub is source-available under the PolyForm Perimeter License 1.0.1, which permits use, modification, and redistribution for permitted purposes while prohibiting competing products. See LICENSING.md for details.
unsloth
unslothai·Python
本地跑和训练大模型、扩散模型的图形界面，支持 GGUF、MLX 及 Qwen、DeepSeek、Gemma、FLUX 等主流模型，省去命令行配置的麻烦。
本地推理+4参数高效微调
79276+1417.4kv0.1.916-beta2026-11-10 11:18 UTC
annotated_deep_learning_paper_implementations
labmlai·Python
用 PyTorch 逐行复现 60 多篇深度学习经典论文，代码旁配注释讲解，涵盖 Transformer、优化器、GAN、强化学习等，适合边读论文边对照实现学习。
论文复现+4参数高效微调
67742+286.8k2026-11-02 05:57 UTC
airllm
lyogavin·Jupyter Notebook
单张 4GB 显存显卡就能跑 70B 甚至 2.8T 参数大模型的分层推理方案，通过逐层加载权重把显存需求压到极低，适合本地低配设备部署大模型。
大模型推理+4参数高效微调
42086+2024.3kv4.0.02026-11-10 06:14 UTC
self-llm
datawhalechina·Jupyter Notebook
面向中文用户的 Linux 环境大模型实战教程，手把手教你在本地微调（全参数/LoRA）和部署国内外开源 LLM 与多模态模型。
大模型微调+4参数高效微调
32744+193.2k2026-10-29 10:59 UTC