Creating a Workflow¶

Step-by-step guide to adding a new workflow to ComfyUI Studio.

Prerequisites¶

Before starting, you need:

A working ComfyUI workflow (tested in the graph editor)
The workflow exported in API format (for static) or broken into logical blocks (for dynamic)
Knowledge of which models and nodes the workflow requires

Choose: Static or Dynamic?¶

If your workflow...	Use
Has a fixed graph structure (same nodes every time)	Static
Was exported from ComfyUI and you just want to parameterize it	Static
Has sections that repeat (N scenes, N batches)	Dynamic
Has conditional sections (enable/disable features)	Dynamic
Needs reusable blocks shared with other workflows	Dynamic

Creating a Static Workflow¶

Step 1: Export from ComfyUI¶

Build and test your workflow in the ComfyUI graph editor
Enable Dev Mode: Settings → Dev Mode → Enable
Click "Save (API Format)" — saves as workflow_api.json
This is your workflow.json

Step 2: Add Placeholders¶

Open the exported JSON and replace hardcoded values with {{variable}} placeholders:

Before:

{"inputs": {"text": "a mountain lake", "seed": 12345, "steps": 20}}

After:

{"inputs": {"text": "{{positive_prompt}}", "seed": "{{seed}}", "steps": "{{steps}}"}}

Every {{name}} must correspond to an input id in the manifest.

Step 3: Create the Directory¶

workflows/my-workflow/
├── manifest.yaml
└── workflow.json

Step 4: Write the Manifest¶

id: my-workflow
category: t2i
name: "My Custom Workflow"
version: 1
date: "2026-03-25 10:00"
type: static
description: "Description of what this workflow does."
author: "Your Name"
workflow_file: workflow.json

inputs:
  - id: positive_prompt
    name: "Prompt"
    type: text
    required: true

  - id: seed
    name: "Seed"
    type: seed
    default: -1

  - id: steps
    name: "Steps"
    type: int
    min: 1
    max: 100
    default: 20

outputs:
  - id: image
    name: "Output Image"
    type: image
    node_id: "15"

required_models:
  - "v1-5-pruned-emaonly.safetensors"

required_nodes:
  - "KSampler"
  - "SaveImage"

Step 5: Add to Index¶

Edit workflows/index.json:

{
  "version": 33,
  "date": "2026-03-25 10:00",
  "workflows": [
    ...existing workflows...,
    {"id": "my-workflow", "version": 1, "date": "2026-03-25 10:00"}
  ]
}

Step 6: Bump Versions¶

Update version.json: - Bump components.workflows.version (integer) - Update components.workflows.date - Update top-level date

Step 7: Test¶

Push to the repository
On the pod, click "Check for Updates"
Go to Workflows — your workflow should appear
Click Run, fill the form, generate

Creating a Dynamic Workflow¶

Step 1: Design the Blocks¶

Break your workflow into logical sections:

Setup block: loads checkpoint, encodes prompts. Exports: model, clip, positive, negative, vae.
Generate block: creates latent, samples, decodes. Imports: model, positive, negative, vae. Exports: image.
Optional blocks: hires fix, face fix, style transfer — each with their own imports/exports.

Step 2: Create Block JSONs¶

For each block, create a JSON file in blocks/:

{
  "nodes": {
    "node_id": {
      "inputs": {"field": "{{variable}}", "connection": ["other_node", 0]},
      "class_type": "NodeClassName",
      "_meta": {"title": "Display Name"}
    }
  },
  "imports": ["model", "positive", "negative", "vae"],
  "exports": {
    "image": {"node": "vae_decode", "output": 0}
  },
  "variables": ["seed", "steps", "cfg"]
}

See Block Format for the complete reference.

Step 3: Define the Pipeline¶

In the manifest:

type: dynamic
blocks_dir: "blocks"
pipeline:
  - block: setup
    file: "setup.json"
  - block: generate
    file: "generate.json"

Step 4: Write Inputs and Complete the Manifest¶

Same as static workflows — define all inputs, outputs, required_models, required_nodes.

Step 5-7: Same as Static¶

Add to index, bump versions, test.

Tips¶

Converting CivitAI Workflows¶

Many workflows from CivitAI are in UI format (not API format). To convert:

Import the workflow into ComfyUI (drag and drop the JSON or image)
Fix any missing nodes (install from ComfyUI-Manager)
Fix model paths (CivitAI workflows often use Windows paths or nested directories)
Test the workflow runs successfully
Export in API format

Model Path Convention¶

ComfyUI expects model filenames relative to the models subdirectory. For example:

Checkpoint: v1-5-pruned-emaonly.safetensors (in models/checkpoints/)
LoRA: my_lora.safetensors (in models/loras/)
VAE: vae-ft-mse-840000-ema-pruned.safetensors (in models/vae/)

Some workflows from CivitAI use subdirectories: WanVideo/2_2/model.safetensors. Make sure your models are downloaded with the correct path structure, or adjust the workflow.

Good tooltips: - Explain what the parameter does in practical terms - Give recommended values for common scenarios - Warn about pitfalls (wrong resolution = bad quality, too high CFG = artifacts) - Reference the checkpoint type when relevant (SD 1.5 vs SDXL vs Pony)

Bad tooltips: - Just repeating the field name ("The seed value") - Technical jargon without explanation - Too short to be useful