This change introduces a complete Local Metadata Builder feature in the resolver
options panel. It allows users to scan local model files, create or update
.metadata.json sidecars, fill missing SHA256 hashes from safetensors headers or
file hashing, and review run history with cancel/error handling.
Changes include:
- backend API: /model_resolver/metadata-build/{capabilities,start,progress/{id},cancel}
- backend worker: core/metadata_builder.py with parallel metadata generation
- frontend section: Local Metadata Builder tab in Options
- frontend polling UI: progress, worker breakdown, history table, status/notifications
Previously, identify_missing_models grouped missing entries only by filename,
which could merge unrelated models from different categories that happen to
share a filename. This change deduplicates by (category, filename) and tracks
all_node_refs with a reference_count so the frontend can preserve one
resolution per node reference instead of collapsing duplicates.
Frontend changes in resolve_download_methods.js iterate over all_node_refs,
and missing_browser_methods.js now shows '+ N refs' in the missing-model
label and clarifies the locate tooltip when multiple workflow references
share the same missing model.
- add POST /model_resolver/local-matches-by-hash backend endpoint
- expand local hash lookup sources to lora_manager_archive, popular, and model_list
- deduplicate backend hash matches by SHA256 value instead of source-hash tuple
- add frontend sync for remote hash matches and hash-label disambiguation
- propagate sha256, file_info, selected_file, and selected_version in model selection/download flows
- render local match status with consistent hash labels across search panel and resolvers
Replace duplicated filename-to-query logic with a single
\uild_filename_search_queries\ helper in matcher/civarchive/huggingface,
standardize Civitai URL generation in the frontend via \getCivitaiModelUrl\,
unify browser user-agent and response file-size extraction in the downloader
path, and derive workflow node-type category hints from widget categories
rather than maintaining a large static mapping.
- Add clear_all_search_caches() in core/sources/__init__.py and replace 4
individual cache-clearing call sites in __init__.py with the new helper.
- Add shared parse_civitai_model_path() in core/type_utils.py and use it
from both CivitAI and CivArchive URL parsers to remove duplicated regex logic.
- Rename parse_huggingface_url() to get_huggingface_repo_and_file_from_url()
in core/resolver.py for clarity.
- Add stripModelExtension() and hasModelExtension() utilities (with a single
MODEL_EXTENSIONS_REGEX) in web/resolver/utils/render_format_methods.js
and replace repeated inline regex patterns across search_panel.js,
model_info_methods.js, tabs_loaded_methods.js, and queue_methods.js.
- Introduce get_filename_from_path in path_utils and replace repeated
os.path.basename calls across core/downloader.py, core/resolver.py,
core/sources/civarchive.py, core/sources/civitai.py, and
core/sources/huggingface.py
- Move duplicated inline HTTP response size helpers (_response_file_size,
_parse_size_header, _parse_content_range_size) into core/type_utils.py
and update civarchive.py and huggingface.py to import them
- Remove inline normalize_sha256_value from __init__.py and reuse the
existing normalize_sha256 helper
- Refactor options_methods.js localStorage read/write logic into a
SETTINGS_MAP-driven loop to reduce repetition
Extract MODEL_EXTENSIONS and format_size_bytes into core/type_utils, removing duplicates from workflow_analyzer and downloader. Add getFilenameFromPath to render_format_methods and replace inline path-splitting logic across JS resolver actions, search, and views.
- Add getHashCompareSourceLabel to normalize source display names
- Add getHashCompareResultName and getHashCompareResultUrl helpers
- Collect and dedupe hash compare matches instead of returning boolean
- Show success notification listing matched sources when local model hash matches online results
Replace duplicated inline drag handlers in queue_methods.js and
missing_browser_methods.js with a new startSplitterDrag() helper in
web/resolver/utils/splitter_drag.js.
The utility centralizes pointer/mouse event binding, RAF throttling,
and touch-cancellation cleanup, cutting net code across the two
callers while preserving existing behaviour (bounds, edge-open
threshold, collapse preview, and settle/cleanup timing).
- Introduce JobProgressTracker in __init__.py to consolidate the duplicated
thread-safe progress/cancellation state and cleanup logic previously
implemented separately for search and hash jobs.
- Extract shared pollHashCalculation method in model_info_methods.js to
centralize hash polling, abort checking, and error handling across all
call sites.
- Use getMissingModelKey in search_panel.js to eliminate inline key
construction duplication.
Implement end-to-end hash comparison for local models and local matches.
Backend:
- Add /model_resolver/local-model-hashes API endpoint
- Add get_local_model_hash_metadata() to read sidecar SHA256 metadata
Frontend:
- Add Compare Hashes context menu item and CSS
- Pass missing_key/missing_search_key through resolveDownloadMethods
- Enhance local match rendering with hash status and context data
- Add hash normalization, formatting, and candidate collection helpers
- Wire compareLocalModelHashesWithCurrentFinding action
Also update HuggingFace author index metadata.
Avoid full list re-render on row selection by passing selectionOnly flag
to displayMissingModels, which now selectively updates row classes and
detail pane instead of rebuilding the entire DOM. Also removed an eager
ensureLocalModelsThenPopulate('') call on combo list presence, and added
a click listener alongside focus so the combo list reliably populates on
both interaction types.
- Added CSS content-visibility and contain-intrinsic-size to list row
containers in resolver-main.css and resolver-shell.css for off-main-thread
rendering optimization.
- Deferred queue list panel split drag updates to requestAnimationFrame in
web/resolver/actions/queue_methods.js, reducing redundant DOM writes and
guarding against duplicate frames during split dragging.
- Deferred missing browser split drag updates to requestAnimationFrame in
web/resolver/views/missing_browser_methods.js with optional chaining on
split start values, preventing layout thrashing while dragging.
Remove manual cursor and pointer-event state restoration during queue
resize. The drag now relies on the browser default col-resize behavior
and avoids setting body styles or mr-is-resizing-queue. Guard
applyQueueSplitWidth against redundant style writes and drop unused
parameters from beginQueueEdgeDrag, deferQueueSplitReleaseUi, and
startSplitDrag.
- Add PointerEvent support for queue toggle, edge drag, and splitter interactions
- Implement deferred queue panel reveal animation (mr-queue-opening class)
- Make queue panel updates demand-driven; skip refresh when panel is hidden
- Replace per-item remove listeners with event delegation on queue list
- Track queue split container width via ResizeObserver with safe fallbacks
- Centralize split flex application and persist queue width/collapsed state with debounce
- Fix missing browser split pane styles to use helper-track CSS variables
- Add deferred UI cleanup for missing browser split drag end
- Add prewarm frame fallback when ResizeObserver is unavailable
- Add /aria2/start and /aria2/stop API endpoints for explicit daemon lifecycle control
- Track aria2 daemon state (managed/running/version/active_transfers) in get_aria2_status
- Implement idle auto-stop timer (5 minutes) with aria2_auto_stop_daemon setting
- Add start_aria2_daemon() and stop_aria2_daemon() helpers in core/downloader.py
- Migrate legacy release-* install directories to normalized versioned paths
- Normalize aria2 version strings (strip release-/v prefixes) in installer metadata
- Update aria2 archive extraction to extract-to-temp then copy into canonical dir
- Add frontend daemon controls: status display, stop button, auto-stop toggle
Dariusz L