Skip to content

Releases: willmiao/ComfyUI-Lora-Manager

v1.2.1

Choose a tag to compare

@willmiao willmiao released this 16 Aug 11:48

Changes

  • Recipes Masonry Layout -- New layout option for the Recipes page that preserves image aspect ratios. Switch between masonry and grid in Settings.
  • Recipe Rematch -- Match recipe models deleted on Civitai against your local library via AutoV3 hashes and re-link them. Available from the right-click context menu for individual recipes, selections, or the whole library.
  • LoRA autocomplete within active filters -- New /activefilters command, with a matching setting in ComfyUI settings, makes LoRA autocomplete in ComfyUI LoRA nodes only suggest LoRAs matching the filters active in the LoRA Manager UI, such as folder, base model, and tags.
  • Drag & Drop Preview -- Set or replace a model card preview by dragging an image or video onto the card. (#1034)
  • Delete Undo -- An Undo toast with a 20 second countdown restores deleted models and recipes.
  • Random sort -- New Random option in the sort dropdowns on model pages and the Recipes page. (#1049)
  • Recipes Recently Opened sort -- Sort recipes by the last time they were opened; never-opened recipes are hidden.
  • Missing model flags in workflow -- LoRA nodes (e.g. Loader, Stacker) now flag LoRAs missing from the local library and block queueing until they're resolved. (#1057)
  • Recipes prompt-aware duplicate detection -- The recipe Duplicates panel now has an Include prompt in matching toggle, so duplicate groups can also require matching prompts.
  • Download failure summary -- Single and batch downloads now show a summary modal with per-item failure reasons, clickable links, Copy Report, and Retry Failed.
  • Metadata Overwrite wired inputs -- The model and sampler fields now accept both manual text entry and wired MODEL/SAMPLER connections, with names extracted automatically when wired.
  • Paid updates detection -- CivitAI versions with paid access are now detected and shown with a Paid badge, plus a new Hide Paid Updates setting and a warning before downloading gated versions. (#1060)

Bug Fixes

  • Fixed Send to Workflow targeting nodes whose text comes from wired inputs, where the send silently did nothing.
  • Fixed updating a model to a version with a different base model saving the file to the old base model's folder. (#1059)
  • Fixed downloads failing with "No suitable file found" for models whose only files use newer CivitAI types (e.g. Enhancement LoRA).
  • Fixed aria2 downloads failing after the download backend restarts mid-transfer - transfers now resume instead of failing.
  • Fixed download URLs without a version ID requiring manual version selection - the latest version is now selected automatically.
  • Fixed autocomplete text widget clear not being undoable - Ctrl+Z now restores cleared text. (#1056)
  • Fixed bulk Set Content Rating not working for selected recipes.
  • Fixed changing the recipes storage path to a different drive failing on Windows.
  • Filter panel improvements - wider panel with height clamped to the viewport, and the 10 preset limit removed.
  • Fixed import modal action buttons scrolling out of view on long lists - they now stay pinned.
  • Example image downloads - Download Missing now skips already-processed models and existing files; Re-process All retries previously failed models.

v1.2.0

Choose a tag to compare

@willmiao willmiao released this 31 Jul 13:26

Changes

  • Create Hook LoRA Node -- New node that outputs selected LoRAs as a hook group compatible with ComfyUI's native hook pipeline, supporting LoRA masking and scheduling.

  • Metadata Overwrite Node -- New node for manually specifying generation parameters when auto-detected metadata is inaccurate or you need more control.

  • Meta Hints User Marks -- Right-click nodes in a workflow and assign a meta hint role (Primary Model, Primary Sampler, prompts) to tell the metadata processor which node to use, overriding heuristic auto-detection.

  • Save Image Node Compression Controls -- New webp_method (compression level) and jpeg_subsampling (chroma quality) node inputs. Each parameter only appears when the corresponding file format is selected.

  • Release/Nightly Channel Switching -- Switch between Release and Nightly update channels and update accordingly.

  • HF Repo Version Grouping -- HF-downloaded models now group by HuggingFace repo in version-group mode (#1040).

Bug Fixes

  • Fixed recipe search freezing the server on libraries with large recipe collections
  • Fixed open Civitai creator profile link not respecting Civitai host setting
  • Fixed AI enrichment failing with LM Studio and other providers that don't support JSON object mode (#1033)
  • Fixed dragging to adjust LoRA strength values on loras node being broken
  • Fixed symlinked library directories being blocked by path containment checks (#1028)
  • Fixed workflow failing to load when LoRA widget received unexpected saved data (#1039)
  • Fixed weighted wildcard syntax (e.g., N::value) not resolving in YAML wildcard files (#1039)
  • Fixed search term being lost when applying or clearing other filters
  • Fixed duplicate model cards appearing when clicking Refresh before download completes (#1041)
  • Fixed recipe card list scrolling to top after downloading missing loras for recipe(s)
  • Fixed delay when opening model modal on network drives

v1.1.9

Choose a tag to compare

@willmiao willmiao released this 21 Jul 14:22

Bug Fixes

  • Fixed recipes page not loading -- a regression from v1.1.8's tag search feature
  • Fixed download path sanitization for empty template placeholders (#1028)
  • Fixed file operations accepting paths outside configured library directories (#1028)

v1.1.8

Choose a tag to compare

@willmiao willmiao released this 20 Jul 12:42

Changes

  • Lora Info Node -- A new display node that shows editable notes and description for a selected LoRA. Connect to a LoRA node (e.g. Loader, Stacker) and selecting a LoRA auto-populates the display.

  • CivArchive API Toggle & Provider Order -- New settings to disable CivArchive API (to avoid rate limits) and to configure the metadata provider fallback order.

  • Tag Search -- Type in the filter panel to search across all tags instead of only browsing the top 20.

  • LoRA Syntax → Path Node -- A new utility node that converts <lora:name:strength> formatted text to the full filesystem path for each LoRA. (#1015)

Bug Fixes

  • Fixed Send to Workflow not working with Primitive nodes and GlobalSeed (#1026)
  • Fixed checkpoints being re-hashed despite having pre-computed hash metadata (#1002)
  • Fixed folder prefix missing from context menu and bulk send-to-workflow when using full path LoRA syntax
  • Fixed bulk mode not exiting after send-to-workflow completes
  • Fixed recipe metadata update bugs -- cache sort crashes and base_model field rejected during validation
  • Fixed download history delete/retry not recognizing the download_id parameter
  • Fixed downloads failing when CivitAI API incorrectly reported models as deleted, causing valid primary download URLs to be skipped -- the primary URL is now used as a fallback when all mirrors are unavailable
  • Fixed standalone mode on python_embeded installations where the script directory was missing from sys.path (#1025)
  • Fixed filter preset tiles not responding when clicking the empty space around the preset name
  • Fixed drop indicator position drifting when the model list is scrolled
  • Fixed Undo/Redo in Vue render mode leaving widget containers detached from the canvas

v1.1.7

Choose a tag to compare

@willmiao willmiao released this 13 Jul 14:19

Changes

  • Batch Hugging Face Downloads -- Paste one or more HF model repo URLs after clicking Download to batch download models from Hugging Face directly.

  • Enrich HF Metadata (AI) -- Right-click a downloaded HF model and select "Enrich HF Metadata (AI)" to fetch base model, tags, trigger words, usage tips, notes, model description, and examples. The metadata is extracted from HF repo content via LLM analysis. deepseek-v4-flash recommended for online, qwen3.5:9b for local.

  • Link to HuggingFace -- Link a local model to a HuggingFace repo, enabling use of Enrich HF Metadata (AI) and View on HuggingFace features.

Bug Fixes

  • Fixed corrupted cache rows breaking all model listings (#730)
  • Fixed user data directories being wiped during Git-based updates
  • Fixed settings data loss when switching portable mode
  • Fixed Send to Workflow not working with expanded subgraph (group) nodes
  • Fixed proxy not applying to aria2 downloads (#1010)
  • Fixed Civitai rate-limited images being permanently removed from metadata
  • Fixed missing models in standalone mode when extra library paths are configured
  • Fixed recipe creation with custom/offline example images (#1012)
  • Fixed license icons showing on models without CivitAI metadata
  • Fixed broken image placeholders for deleted preview files
  • Fixed LoRA nodes and TriggerWord Toggle node auto-resizing -- these nodes now maintain their size regardless of content changes, works on both legacy mode and Node 2.0
  • Fixed overlapping checkpoint/UNet folder paths being silently accepted

v1.1.6

Choose a tag to compare

@willmiao willmiao released this 28 Jun 14:17

Bug Fixes

  • Aria2 download stability -- Fixed large model downloads failing in Docker environments.
  • Fixed Updates page showing false update notifications for models already in other folders (#997) -- The Updates page no longer shows versions as available for download when they already exist in your library but in a different folder. Folder filtering now checks the entire library.
  • Fixed recipe import not detecting LoRAs when CivitAI metadata is null -- Recipes imported from CivitAI images now properly discover LoRAs and checkpoints even when the API returns incomplete metadata.
  • Fixed download routing for UNet and Diffusion Model file types -- Checkpoint downloads now correctly route UNet and Diffusion Model files to the proper download folder.
  • Fixed marquee selection breaking when scrolling -- Bulk selection marquee no longer deselects cards when scrolling during a drag, with auto-scroll when dragging near viewport edges.
  • Fixed Send to Workflow node detection with multiple ComfyUI tabs open -- Workflow nodes from different tabs are now correctly merged instead of overwriting each other, preventing Send to Workflow failures.
  • Fixed Launch LoRA Manager button disappearing in subgraphs (#996) -- The button no longer disappears when opening the properties panel in a ComfyUI subgraph.
  • Fixed trigger word strength wheel not working in Vue render mode -- Mouse wheel strength adjustment on trigger words now works in ComfyUI's Vue render mode.
  • Custom sort dropdown -- Replaced the native browser sort select with a custom dropdown. Sort selections now persist per page mode and recipes remember their sort order.
  • Fixed bulk-mode highlight being clipped on edge cards -- Batch mode selection highlights are no longer clipped on cards at the grid edges.

v1.1.5

Choose a tag to compare

@willmiao willmiao released this 24 Jun 14:11

Changes

  • Group by Model -- Collapse multiple Civitai versions of the same model into a single card. Toggle from Settings or the global right-click context menu. Grouped cards show a clickable version count; clicking opens View Local Versions.

  • Version Grouping -- Renamed from Update Flag Strategy. Controls how versions are grouped and how updates are flagged: group by model and base model, or merge all versions together.

  • Local Versions sort -- Sort by version count. Works in both group-by-model mode (sorts grouped cards) and normal mode (sorts individual version cards).

  • Send Prompt and Gen Params to Workflow -- Mark ComfyUI nodes as targets via right-click, then send prompts and generation params (seed, steps, sampler, scheduler, CFG) from the recipe modal or metadata panel.

  • View all local versions -- The button in the model details tab is now functional, showing all local versions of a model on the main page.

Bug Fixes

  • %batch_num% in Save Image node -- Use the %batch_num% placeholder to include the batch loop index in saved filenames.
  • Search bar keyboard shortcut cue -- Ctrl+F shortcut now shows a visual hint. Clear button positioning fixed.
  • Fixed full-page loading flicker -- Loading overlay now scoped to the card grid.

v1.1.4

Choose a tag to compare

@willmiao willmiao released this 19 Jun 10:31

Changes

  • Theme presets -- five built-in color schemes to choose from.
  • Updated license icons -- new icons matching the current CivitAI design with colored permission indicators; toggleable via the Use updated license icons setting.
  • API key now hidden from frontend -- settings shows Configured status instead of exposing the key value; save-password prompts prevented in Chromium and Firefox.
  • Statistics page fully internationalized -- all content including chart labels, tooltips, and insights now support your UI language. (#489)
  • Model Types chart in the Collection tab on the Statistics page.

Bug Fixes

  • Fixed tag case sensitivity -- tags are normalized to lowercase on save; filtering is now case-insensitive. (#727)
  • Fixed LoraTextLoaderLM metadata -- SaveImageLM now correctly records LoRAs loaded via LoraTextLoaderLM. (#801)
  • Fixed Base Model Distribution chart sorting -- X-axis labels now sorted alphabetically. (#796)
  • Fixed example images validation -- hidden/system files (.DS_Store, Thumbs.db) are automatically skipped; error messages list offending filenames. (#807)
  • Fixed Duplicate Manager theme colors -- now correctly respects the active color theme.
  • Unified recipe modal tag UI -- recipe tag display and editing now use the same shared components as model tags, fixing cross-modal conflicts.
  • Fixed drag-to-folder causing grid reload -- scroll position is now preserved when moving models between folders, keeping the page state for uninterrupted operations.
  • Improved sidebar restore indicator -- repositioned to align with the sidebar header with a pulsing animation for better discoverability on first use. (#990)
  • Batch import summary modal visual cleanup -- toned-down styling consistent with the rest of the UI.

v1.1.3

Choose a tag to compare

@willmiao willmiao released this 17 Jun 09:53

Changes

  • Card Overlay Blur slider redesign -- now matches the settings UI style.

Bug Fixes

  • Fixed error when scanning new LoRAs on Windows -- SHA256 hash computation threw an exception, causing newly added .safetensors files to be skipped as invalid. (#988)

v1.1.2

Choose a tag to compare

@willmiao willmiao released this 17 Jun 01:58

Changes

  • AVIF and JXL example image support -- AVIF and JXL images are now supported as example images, with metadata extraction from their ISOBMFF brotli-compressed containers.
  • %NodeTitle.WidgetName% placeholders in Save Image node -- filename_prefix can now reference widget values from other nodes using %NodeTitle.WidgetName% patterns, alongside %date:format% and other standard placeholders. (#314)
  • Metadata fetch result summary modal -- bulk metadata refresh now shows a detailed summary with success, failure, and skip counts, along with individual error details. (#38)
  • Folder sidebar simplified to per-page toggle -- removed the global folder sidebar visibility setting and auto-hide hover mechanism. The folder sidebar is now hidden/shown per-page via chevron buttons in the header. Old settings are auto-migrated.

Bug Fixes

  • Fixed bulk operations not stopping immediately on cancel -- canceling during scan, update check, or bulk delete now actually halts the backend, with a Cancelling... status indicator. (#986)
  • Fixed models marked as not-found re-fetching on every restart -- the not found status now persists to the SQLite cache even when the model was classified through the CivArchive/SQLite provider path.
  • Fixed high memory usage when hashing large checkpoint files -- SHA256 computation now uses posix_fadvise to avoid polluting the OS page cache. Lazy hash is properly respected for checkpoint scans at startup, and scanner tasks are cancelled on shutdown to prevent WSL hangs. (#985)
  • Fixed CivArchive rate limits freezing bulk metadata refresh -- rate-limited providers are now skipped to the next fallback, and the operation auto-aborts after 3 consecutive rate limits with a clean message. (#983)