The model field now accepts either a manual string or a MODEL connection.
When wired, the model name is extracted from the patcher's
cached_patcher_init (registered by core loaders load_checkpoint_guess_config
and load_diffusion_model, preserved through LoRA clones) and converted to a
ComfyUI-style relative name via config model roots.
- model input declared as "STRING,MODEL" with widgetType STRING, so the
text widget and the dual-type connection slot coexist; non-STRING/MODEL
links are rejected by frontend and backend type validation
- UNETLoaderLM GGUF branch now registers a custom cached_patcher_init reload
factory so GGUF models participate in name extraction and ModelPatcher
deepclone/dynamic machinery
- shared collect_overwrite_params() helper keeps the node and the metadata
extractor conversion logic in sync; extraction failures are logged instead
of silently dropping the overwrite
Two bugs prevented type-signature-based fallback from working:
- metadata_hook.py used getattr(obj.__class__, 'RETURN_TYPES')
which fails when _async_map_node_over_list is called with
a class (not instance) — obj.__class__ is the metaclass
'type', which has no RETURN_TYPES. Fixed: getattr(obj, ...).
- metadata_registry.py used type(extractor) is GenericNodeExtractor
to dispatch return_types. NODE_EXTRACTORS stores class
references, not instances; type(Class) is always 'type',
never the class. Fixed: extractor is GenericNodeExtractor.
GenericNodeExtractor (previously a no-op) now inspects
RETURN_TYPES to detect MODEL loaders and CONDITIONING
encoders in nodes not registered in NODE_EXTRACTORS.
- Propagate return_types from the hook layer through the
registry to GenericNodeExtractor.extract() and update().
- MODEL detection: scan ckpt_name/unet_name/model_path/
model_name/gguf_name fields, validate by extension.
- CONDITIONING detection: scan text/clip_l/t5xxl/prompt
fields, store prompt text and conditioning tensor.
- _fill_missing_metadata also checks node_cache, so
GenericNodeExtractor-handled nodes survive cache.
Users can now right-click nodes and assign meta hints
(primary_model, primary_sampler, positive_prompt,
negative_prompt) to override the metadata processor's
heuristic inference.
- Store extra_data from the API request so workflow node
properties (including lm_marker_role) are accessible
during metadata processing.
- _get_user_marks scans extra_data.extra_pnginfo.workflow
for meta_* marks, falling back to prompt.original_prompt.
- extract_generation_params checks user marks before
heuristic inference for sampler, model, and prompts.
- Warn on duplicate marks or invalid marked nodes.
Fixes#394 — LoRAs loaded via rgthree Power Lora Loader were not
tracked in usage statistics because no extractor existed for that node.
New extractors:
- RgthreePowerLoraLoaderExtractor: parses LORA_* kwargs, respects
the per-LoRA 'on' toggle
- TensorRTLoaderExtractor: parses engine filename (strips _$profile
suffix) as best-effort for vanilla TRT. If the output MODEL has
attachments["source_model"] (set by NubeBuster fork), overrides
with the real checkpoint name.
TensorRTRefitLoader and TensorRTLoaderAuto take a MODEL input whose
upstream checkpoint loader is already tracked — no extractor needed.
Also adds a name:<filename> fallback and warning log in both
_process_checkpoints and _process_loras when hash lookup fails.
Fix issue #866 where the metadata hook's async wrapper used *args/**kwargs
which caused AttributeError when ComfyUI's make_locked_method_func tried
to access __func__ on the func parameter.
The async_map_node_over_list_with_metadata wrapper now uses the exact
same signature as ComfyUI's _async_map_node_over_list:
- Removed: *args, **kwargs
- Added: explicit v3_data=None parameter
This ensures the func parameter (always a string like obj.FUNCTION) is
passed correctly to make_locked_method_func without any type conversion.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add pre-processing step to populate missing parameters for candidate samplers, especially for SamplerCustomAdvanced requiring tracing
- Change sampler selection from most recent (closest to downstream) to first in execution order to prioritize base samplers over refine samplers
- Improve parameter handling by updating sampler parameters with traced values before ranking
- Maintain backward compatibility with fallback to first sampler if no criteria match
- Add support for `basic_pipe` nodes in metadata processor to handle pipeline nodes like FromBasicPipe
- Optimize `find_primary_checkpoint` by accepting optional `primary_sampler_id` to avoid redundant calculations
- Update `get_workflow_trace` to pass known primary sampler ID for improved efficiency
Update metadata registry to remove cache entries when node metadata becomes empty instead of keeping stale data. This prevents accumulation of unused cache entries and ensures cache only contains valid metadata. Added test case to verify cache behavior when LoRA configurations are removed.
- Add CheckpointLoaderKJ to NODE_EXTRACTORS mapping for KJNodes support
- Enhance model filename generation in SaveImage to handle different data types
- Add proper type checking and fallback for model metadata values
- Improve robustness when processing checkpoint paths for filename generation
Add KJNodesModelLoaderExtractor to handle metadata extraction from KJNodes loaders that expose model_name. This supports GGUFLoaderKJ and DiffusionModelLoaderKJ nodes, ensuring consistent checkpoint metadata collection across different node types.
Add GGUFLoaderExtractor class to handle metadata extraction for GGUF model loaders. Register extractor for both LoaderGGUF and LoaderGGUFAdvanced node types to capture checkpoint metadata from gguf_name input parameter. This enables proper metadata tracking for GGUF model files used in the system.
