workflow

name	workflow
description	Create a ComfyUI workflow from a template

Create Workflow

Create and optionally run a ComfyUI workflow. Template or description provided by the user as input: "$ARGUMENTS".

Steps

1. Pick a template. If no specific template was requested, choose from the built-in template names accepted by comfyui_create_workflow: txt2img, img2img, upscale, inpaint, txt2vid_animatediff, txt2vid_wan, controlnet_canny, controlnet_depth, controlnet_openpose, ip_adapter, lora_stack, face_restore, flux_txt2img, sdxl_txt2img. The full list is also in comfyui_create_workflow's docstring. (Note: comfyui_list_workflows returns server-side /workflow_templates from front-end packages — not these MCP-built-in templates.) If the user provides a template name that is not in this list, respond with an error message and list the valid options above.

2. Create the workflow.

   • If the user did not specify any parameter overrides, call comfyui_create_workflow with only the template name (omit params; it defaults to "", which applies the predefined settings for the selected template).
   • If the user specified overrides (e.g., prompt, dimensions), pass them as a JSON string in the params argument:

     create_workflow(template="txt2img", params='{"prompt": "a sunset", "width": 768}')
     

3. Validate the workflow. Call comfyui_validate_workflow with the workflow JSON returned from step 2. The response is a dict {valid, errors, warnings, node_count, pipeline}. Report any entries in errors (blocking) and warnings (non-blocking).

4. Present the workflow. Show a summary of what the workflow does (nodes, connections, key parameters). Offer the user two options:

   • Run it — call comfyui_run_workflow with wait=True. The response is a dict envelope: {status, prompt_id, outputs, elapsed_seconds, warnings?}. Read status to confirm completion.
   • Modify it — use comfyui_modify_workflow to add/remove nodes or change parameters, then re-validate

Notes

• Use comfyui_summarize_workflow for a human-readable text or Mermaid overview (output_format="text" or "mermaid").
• Use comfyui_analyze_workflow if you need the structured analysis as a dict (fields like pipeline, models, parameters) without parsing prose.
• For simple text-to-image generation, /comfy:gen is faster.