Format node type by removing the last underscore and everything after it. Also handles node names with parentheses (e.g., "Node Name (author)

Resolve a model path relative to ComfyUI's model directories. Args: model_path: Model path (can be relative or absolute) mod

Get the base path of the ComfyUI installation. Returns: Path to ComfyUI base directory or None if not found

Fetch metadata information for a node type. Args: node_type: Node type name Returns: Node metadata

Get full path to a model using ComfyUI's folder_paths. Args: model_name: Model name (e.g., "workflow-generator-q8_0.gguf" or "Qwen2.

Find the k most similar nodes to the given node type using semantic search. Args: node_type: Node type name to search fo

Build configuration for NodeBase. Args: catalog_directory: Directory containing node_list.json and node_info.json

Auto-detect model format from model name. Args: model_name: Model name (e.g., "workflow-generator-q8_0.gguf" or "Qwen2.5-7B-Instruct

Register "llm" folder type in ComfyUI's folder_paths. This allows us to use folder_paths.get_filename_list("llm") and folder_paths.get_full_p

Resolve a configuration path relative to the custom node directory or ComfyUI root. Args: path: Path string (can be relative or abso

Recursively scan directory for .gguf files and HuggingFace model directories. Args: directory: Directory to scan

Automatically detect tokenizer path based on model path. For GGUF models: Tokenizer is typically in a directory with the same name as th

Build configuration for NodeValidator. Args: local: Whether to use local model model_format: Model format ("

Build configuration for WorkflowGenerator. Args: model_format: Model format ("gguf" or "huggingface") model_

Extract input names from node meta info. Args: meta_info: Node metadata dictionary containing 'inputs' key Returns: Lis

Fix node name using semantic search only (non-local mode). Args: desc: User description diagram: Workflow di

Fix node name using local language model. Args: desc: User description diagram: Workflow diagram

Generate workflow diagram from natural language prompt. Args: prompt: Natural language description of the desired workfl

Get the path to the ComfyUI-WorkflowGenerator custom node directory. Returns: Path to custom node directory

Get all models from ComfyUI/models/LLM/ (both .gguf files and directories). Recursively scans subdirectories for .gguf files. Returns onl

Determine which lazy inputs need to be evaluated for LLM refinement. Handles both standard parameter names (NodeValidator) and prefixed names

Clean up model resources.

Clean up model resources.

Extract information from a V1 API node. Args: node_name: Node class name node_class: Node class object Returns:

Extract output names from node meta info. Args: meta_info: Node metadata dictionary containing 'output_names' or 'outputs' key

Get device string for model loading. Args: device_preference: "auto", "cuda", or "cpu" Returns: Device string ("cuda" o

Get dtype for HuggingFace model loading. Args: dtype_preference: "auto", "fp16", "bf16", "fp32", or "fp8" device: Device str

Filter embedding models (SentenceTransformer/HuggingFace directories only). Excludes .gguf files as embedding models are typically HuggingFac

Generate options for n_threads Combo input. Returns: List of strings ["All", "Auto"] + integer options up to CPU count

Filter models suitable for refine_agent (node validator). Includes both GGUF files and HuggingFace directories. Returns: List of

Filter models suitable for workflow generator. Includes both GGUF files and HuggingFace directories. Returns: List of model name

Check if a node type is a valid ComfyUI node class name. This checks against ComfyUI's actual NODE_CLASS_MAPPINGS, not just the cata

Unified model loader that supports both GGUF and HuggingFace formats. Args: model_format: "gguf" or "huggingface" model_path

Check if a node type exists in the node list. Args: node_type: Node type name to check Returns: Tru

Parse a workflow diagram into ComfyUI workflow JSON. Args: diagram: List of edges representing the workflow diagram

Parse n_threads Combo value to integer or None. Args: value: String from Combo input ("All", "Auto", or integer string) Returns

Refine a workflow diagram by validating and correcting node names. Args: diagram: List of edges representing the workflo

Save workflow JSON to a file with smart incrementing counter. Args: workflow_json: The workflow JSON string to save. filenam

Calculate and apply positions to all nodes. Modifies the 'pos' attribute of nodes in-place.

Calculate optimal number of GPU layers based on available VRAM. Args: model_size_gb: Total model size in GB available_vram_g

Decode an edge and create a link between source and destination nodes. Args: outputs: List of output dictionaries from source node

Remove primitive nodes from diagram. Args: digram: List of edges in the diagram, where each edge is [src_type, src_name, dst_type, d

Extract information from a V3 API node. Args: node_name: Node class name node_class: Node class object (io.ComfyNode) R

Fetch input information for a node type. Args: node_type: Node type name Returns: Tuple of (inputs

Fetch output information for a node type. Args: node_type: Node type name Returns: List of output d

Get available VRAM in GB using ComfyUI's memory management if available. Returns: Available VRAM in GB

Get candidate inputs that can be added dynamically. Args: input_types: Dictionary of input type definitions inputs: List of

Get diagram information for a node type with enhanced metadata. Args: node_type: Node type name Returns:

Extract output name to index mapping from outputs list. Args: outputs: List of output dictionaries Returns: Dictionary

Extract and parse JSON from encoded string. Args: encode: String containing JSON (possibly wrapped in markdown code blocks) or alrea

Post-process the model response into a valid diagram format. Args: response: Raw model response (string or list)

Static convenience method

Scan all available ComfyUI nodes (V1 and V3). Args: progress_callback: Optional callback function(current, total) to report progress

Update the node catalog by scanning ComfyUI nodes. Args: node_list_path: Path to save node list JSON node_info_db_path: Path

Cleanup method for VRAM management.

Cleanup method for VRAM management.

Initialize layout engine. Args: nodes: List of node dictionaries links: List of link arrays [id, src_id, src

Initialize GGUF model wrapper. Args: model_path: Path to GGUF model file tokenizer_path: Path to HuggingFace

Initialize HuggingFace model wrapper. Args: model_path: Path to HuggingFace model directory tokenizer_path:

Initialize WorkflowGenerator with model configuration. Args: cfg: Configuration object with workflow_generator settings

Initialize WorkflowBuilder. Args: node_base: NodeBase instance for node information

Initialize NodeBase with node database and optionally embedding model. Args: cfg: Configuration object with node_base se

Initialize NodeValidator. Args: node_base: NodeBase instance for node information cfg: Configuration object

Initialize ConfigBuilder. Args: base_config: Base configuration dictionary (default: DEFAULT_CONFIG)

Build full configuration for all generators. Args: workflow_generator_kwargs: WorkflowGenerator configuration overrides

Determine which lazy inputs need to be evaluated. Returns: List of input names that need to be evaluated. LL

Determine which lazy inputs need to be evaluated. Returns: List of input names that need to be evaluated. Re

Entry point for ComfyUI to load the extension.

Execute node catalog update.

Execute NodeValidator to validate and correct node names.

Execute complete ComfyUI-WorkflowGenerator pipeline.

Execute WorkflowBuilder to convert diagram to workflow. Uses internal caching to avoid rebuilding workflows when only save settings

Execute WorkflowGenerator to generate workflow diagram.

Fingerprint inputs to determine if node should re-execute. Returns: Hashable value that changes when node should re-exec

Fingerprint inputs to determine if node should re-execute. Returns: Tuple of relevant inputs (no hashing needed - ComfyU

Fingerprint inputs to determine if node should re-execute. Returns: Tuple of relevant inputs (no hashing needed - ComfyU

Fingerprint inputs to determine if node should re-execute. Caching strategy: - Only includes inputs that affect the workflow

Fingerprint inputs to determine if node should re-execute. Returns: Tuple of relevant inputs (no hashing needed - ComfyU

Generate text using GGUF model. Args: input_ids: Token IDs from HuggingFace tokenizer (tensor or list) max_n

Generate text using HuggingFace model. Args: input_ids: Token IDs tensor max_new_tokens: Maximum number of t

Get device information using ComfyUI's model management. Returns: Tuple of (device_string, cuda_available)

Return list of ComfyUI-WorkflowGenerator node classes. Returns: List of node classes to register

Test that NodeValidator node has correct schema.

Test that Pipeline node has correct schema.

Test that UpdateNodeCatalog node has correct schema.

Test that WorkflowBuilder node has correct schema.

Test that WorkflowGenerator node has correct schema.

Validate inputs before execution. Returns: True if valid, or error message string if invalid

Validate constant inputs before execution. Note: Only constant inputs (widget values) are available here. Connected inputs (

Validate constant inputs before execution. Note: Only constant inputs (widget values) are available here. Connected inputs a

Validate constant inputs before execution. Note: Only constant inputs (widget values) are available here. Connected inputs (

Validate constant inputs before execution. Note: Only constant inputs (widget values) are available here. Connected inputs a