ComfyUI Workflow Generator

A Python package for generating ComfyUI workflow APIs from object_info.json using AST code generator. Like ComfyScript, but much more simple.

Installation

pip install comfyui-workflow-generator

Usage

1. Generate Workflow API

From Command Line

# Generate from local object_info.json (auto-detected)
comfyui-generate object_info.json -o my_workflow_api.py

# Generate from ComfyUI server (auto-detected)
comfyui-generate http://127.0.0.1:8188 -o workflow_api.py

# Generate with default output name
comfyui-generate object_info.json

From Python

# Import the generated API
from workflow_api import Workflow
from comfyui_workflow_generator import ComfyUIWorkflowExecutor

# Step 1: Upload image with custom name
executor = ComfyUIWorkflowExecutor("http://127.0.0.1:8188")
uploaded_filename = executor.upload_image(
    "my_very_own_image.png", 
    image_name="my_very_own_image.png"
)

# Step 2: Create workflow
wf = Workflow()

# Load checkpoint
model, clip, _ = wf.CheckpointLoaderSimple(ckpt_name="Illustrious-XL-v1.0.safetensors")

# Load vae, i usually do it via separate node
vae = wf.VAELoader(vae_name="sdxl_vae.safetensors")

# Load uploaded image (use the uploaded filename)
image, mask = wf.LoadImage(image="my_very_own_image.png")

# Encode prompts
positive = wf.CLIPTextEncode(text="1girl, lisa_\(genshin_impact\)", clip=clip)
negative = wf.CLIPTextEncode(text="blurry, low quality", clip=clip)

# Encode image for img2img
latent = wf.VAEEncode(pixels=image, vae=vae)

# Sample
denoized_latent = wf.KSampler(
    model=model,
    seed=42,
    steps=20,
    cfg=8.0,
    sampler_name="euler",
    scheduler="normal",
    positive=positive,
    negative=negative,
    latent_image=latent,
    denoise=0.8
)

# Decode
result_image = wf.VAEDecode(samples=denoized_latent, vae=vae)

# Save
wf.SaveImage(images=result_image, filename_prefix="lisa_from_genshin")

# Step 3: Get workflow JSON (for debugging or manual inspection)
workflow_json = wf.get_workflow()
print("Generated workflow:", workflow_json)

# save it to file, it could be loaded in comfyui
with open("workflow_json.json", "w") as f:
    f.write(workflow_json)

# Step 4: Execute it (do not feed it with json, it expects dict)
results = executor.execute_workflow(wf.workflow_dict, output_dir="./results")

Generated workflow look like this:

Key Design Principles

1. No Input Manipulation

The workflow executor executes workflows exactly as provided, without any modifications.

1. No Weird Features

No nested event loops, no Real Mode, Unreal Mode, or other terms from the jargon of x86 assembler programmers of the 90s. It's just a workflow generator, thicc as brick.

3. Type Safety

Generated APIs include proper return type hints:

def CheckpointLoaderSimple(self, ckpt_name: str) ->(MODEL, CLIP, VAE):
    ...

def LoadImage(self, image: str) ->(IMAGE, MASK):
    ...

def KSampler(self, model: MODEL, seed: int, steps: int, cfg: float,
        sampler_name: str, scheduler: str, positive: CONDITIONING, negative:
        CONDITIONING, latent_image: LATENT, denoise: float) ->LATENT:
    ...

Package Structure

comfyui_workflow_generator/
├── __init__.py          # Package exports
├── generator.py         # AST-based code generator
├── executor.py          # Workflow execution (no input manipulation)
└── cli.py              # Command-line interface

Development

# Install in development mode
pip install -e .

# Run tests
pytest

# Format code
black comfyui_workflow_generator/

# Lint code
flake8 comfyui_workflow_generator/