ComfyUI Node
Workflow Generator Pipeline
Complete Pipeline - Executes steps 1-3 sequentially: 1. WorkflowGenerator -> Creates initial diagram 2. NodeValidator -> Validates/corrects node names 3. Build and Save -> Converts to JSON Convenience node for the full process.
Workflow Generator Pipeline
- workflow_edges
- workflow_edges (refined)
- workflow_json
- file_path
◄instruction—►
◄model_pathworkflow-generator-q8_0.gguf►
◄dtypeauto►
◄device_preferenceauto►
◄attn_implementationauto►
◄auto_gpu_layerstrue►
◄n_gpu_layers-1►
◄max_new_tokens8192►
◄context_size4096►
◄temperature0.95►
◄top_p0.70►
◄allow_primitive_nodesfalse►
◄refine_enabledtrue►
◄use_llm_refinementfalse►
◄refine_model_pathQwen2.5-7B-Instruct-q8_0.gguf►
◄embedding_model_pathparaphrase-multilingual-MiniLM-L12-v2►
◄catalog_directorycatalog►
◄refine_dtypeauto►
◄refine_device_preferenceauto►
◄refine_attn_implementationauto►
◄refine_auto_gpu_layersfalse►
◄refine_n_gpu_layers-1►
◄refine_max_new_tokens4096►
◄refine_context_size4096►
◄refine_temperature0.95►
◄top_k5►
◄refine_top_p0.70►
◄save_workflowtrue►
◄filename_prefixgenerated_workflow►
◄seed0►
CategoryWorkflowGenerator
Inputs (30)
| Name | Type | Default | Description |
|---|---|---|---|
| instruction | STRING | Description of the desired ComfyUI workflow. | |
| model_path | COMBO | workflow-generator-q8_0.gguf | Model file (GGUF) or directory (HuggingFace). |
| dtype | COMBO | auto | Data type (HuggingFace only). |
| device_preference | COMBO | auto | Device preference (cuda/cpu). |
| attn_implementation | COMBO | auto | Attention implementation (HuggingFace). |
| auto_gpu_layers | BOOLEAN | true | Auto-calculate GPU layers based on available VRAM (prevents OOM errors) (GGUF). |
| n_gpu_layers | INT | -1-1–1000 | Number of GPU layers (GGUF). -1 for all. |
| max_new_tokens | INT | 81921–32768 | Max tokens to generate. |
| context_size | INT | 4096512–131072 | Context window size (GGUF). |
| temperature | FLOAT | 0.950–2 | Sampling temperature. |
| top_p | FLOAT | 0.700–1 | Top-p sampling. |
| allow_primitive_nodes | BOOLEAN | false | Include primitive nodes in diagram. False removes them for cleaner workflows. |
| refine_enabled | BOOLEAN | true | Enable Step 2 (NodeValidator). |
| use_llm_refinement | BOOLEAN | false | Enable LLM-based node correction (slower but more accurate). |
| refine_model_path | COMBO | Qwen2.5-7B-Instruct-q8_0.gguf | LLM model for refinement. |
| embedding_model_path | COMBO | paraphrase-multilingual-MiniLM-L12-v2 | Embedding model for semantic search. |
| catalog_directory | STRING | catalog | Directory containing node catalog files. |
| refine_dtype | COMBO | auto | Data type for refinement model (HuggingFace). |
| refine_device_preference | COMBO | auto | Device preference for refinement model. |
| refine_attn_implementation | COMBO | auto | Attention implementation for refinement model. |
| refine_auto_gpu_layers | BOOLEAN | false | Auto-calculate GPU layers for refinement model (GGUF). |
| refine_n_gpu_layers | INT | -1-1–1000 | GPU layers for refinement model (GGUF). -1 for all. |
| refine_max_new_tokens | INT | 40961–16384 | Max tokens for refinement. |
| refine_context_size | INT | 4096512–32768 | Context window size for refinement model (GGUF). |
| refine_temperature | FLOAT | 0.950–2 | Temperature for refinement. |
| top_k | INT | 51–20 | Number of similar nodes to consider. |
| refine_top_p | FLOAT | 0.700–1 | Top-p for refinement. |
| save_workflow | BOOLEAN | true | Save workflow to file. |
| filename_prefix | STRING | generated_workflow | Filename prefix or path for saved workflow. |
| seed | INT | 00–4294967295 | Random seed. |
Outputs (4)
| Name | Type | Description |
|---|---|---|
| workflow_edges | STRING | Initial workflow diagram (JSON). |
| workflow_edges (refined) | STRING | Refined workflow diagram (JSON). |
| workflow_json | STRING | Final ComfyUI workflow JSON. |
| file_path | STRING | Path to saved file. |