ComfyUI Cheatsheet: Nodes, Workflows and API Reference

# Clone + Python venv install
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
python -m venv venv && source venv/bin/activate

# NVIDIA (CUDA 12)
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
pip install -r requirements.txt

# Apple Silicon — Metal backend
pip install torch torchvision torchaudio          # nightly-MPS builds are usually fine
pip install -r requirements.txt

# Drop checkpoints / LoRAs / VAEs into:
#   models/checkpoints/   models/loras/   models/vae/   models/controlnet/

# Launch (web UI at http://127.0.0.1:8188)
python main.py
# Listen on LAN:        python main.py --listen 0.0.0.0
# Low VRAM (<6GB):      python main.py --lowvram
# CPU only (slow):      python main.py --cpu
# Auto-launch browser:  python main.py --auto-launch

{
  "3": {
    "class_type": "KSampler",
    "inputs": {
      "seed": 42,
      "steps": 25,
      "cfg": 7.0,
      "sampler_name": "dpmpp_2m",
      "scheduler": "karras",
      "denoise": 1.0,
      "model":           ["4", 0],
      "positive":        ["6", 0],
      "negative":        ["7", 0],
      "latent_image":    ["5", 0]
    }
  },
  "4": {
    "class_type": "CheckpointLoaderSimple",
    "inputs": { "ckpt_name": "sd_xl_base_1.0.safetensors" }
  },
  "5": {
    "class_type": "EmptyLatentImage",
    "inputs": { "width": 1024, "height": 1024, "batch_size": 1 }
  },
  "6": {
    "class_type": "CLIPTextEncode",
    "inputs": { "text": "cinematic photo of an old fisherman", "clip": ["4", 1] }
  },
  "7": {
    "class_type": "CLIPTextEncode",
    "inputs": { "text": "blurry, text, watermark",            "clip": ["4", 1] }
  },
  "8": {
    "class_type": "VAEDecode",
    "inputs": { "samples": ["3", 0], "vae": ["4", 2] }
  },
  "9": {
    "class_type": "SaveImage",
    "inputs": { "filename_prefix": "out", "images": ["8", 0] }
  }
}

import json, urllib.request, uuid

SERVER = "127.0.0.1:8188"
CLIENT_ID = str(uuid.uuid4())

# 1 · Load a workflow exported as API JSON (Settings → "Save (API Format)")
with open("workflow_api.json") as f:
    workflow = json.load(f)

# 2 · Override values just like editing nodes in the UI
workflow["6"]["inputs"]["text"] = "a wide-angle photo of a foggy harbour at dawn"
workflow["3"]["inputs"]["seed"] = 123_456

# 3 · Queue the prompt
req = urllib.request.Request(
    f"http://{SERVER}/prompt",
    data=json.dumps({"prompt": workflow, "client_id": CLIENT_ID}).encode(),
    headers={"Content-Type": "application/json"},
)
prompt_id = json.loads(urllib.request.urlopen(req).read())["prompt_id"]
print("queued:", prompt_id)

# 4 · Poll history → get the saved file
hist = json.loads(
    urllib.request.urlopen(f"http://{SERVER}/history/{prompt_id}").read()
)[prompt_id]
for node_out in hist["outputs"].values():
    for img in node_out.get("images", []):
        url = f"http://{SERVER}/view?filename={img['filename']}&subfolder={img['subfolder']}&type={img['type']}"
        print("image:", url)

# custom_nodes/my_pack/__init__.py
class GrayscaleNode:
    """Convert an image batch to grayscale."""

    @classmethod
    def INPUT_TYPES(cls):
        return {
            "required": {
                "image":    ("IMAGE",),
                "strength": ("FLOAT", {"default": 1.0, "min": 0.0, "max": 1.0, "step": 0.01}),
            },
        }

    RETURN_TYPES  = ("IMAGE",)
    RETURN_NAMES  = ("image",)
    FUNCTION      = "run"
    CATEGORY      = "image/filters"

    def run(self, image, strength):
        # image is a torch.Tensor of shape [B, H, W, C], values in 0..1
        gray = image.mean(dim=-1, keepdim=True).repeat(1, 1, 1, 3)
        return (gray * strength + image * (1 - strength),)

NODE_CLASS_MAPPINGS = {"Grayscale": GrayscaleNode}
NODE_DISPLAY_NAME_MAPPINGS = {"Grayscale": "Grayscale (mine)"}

import json, uuid, urllib.parse, urllib.request, websocket

SERVER = "127.0.0.1:8188"
CLIENT_ID = str(uuid.uuid4())

with open("workflow_api.json") as f:
    workflow = json.load(f)
workflow["6"]["inputs"]["text"] = "a tea stall in monsoon Mumbai, cinematic, 35mm"

# Subscribe to progress BEFORE queuing the prompt
ws = websocket.WebSocket()
ws.connect(f"ws://{SERVER}/ws?clientId={CLIENT_ID}")

req = urllib.request.Request(
    f"http://{SERVER}/prompt",
    data=json.dumps({"prompt": workflow, "client_id": CLIENT_ID}).encode(),
    headers={"Content-Type": "application/json"},
)
prompt_id = json.loads(urllib.request.urlopen(req).read())["prompt_id"]

# Stream events: progress %, current node, completion
while True:
    msg = ws.recv()
    if not isinstance(msg, str):
        continue
    event = json.loads(msg)
    data = event["data"]
    if event["type"] == "progress":
        print(f"step {data['value']} / {data['max']}")
    elif event["type"] == "executing" and data["node"] is None and data["prompt_id"] == prompt_id:
        break       # done

# Fetch the result
hist = json.loads(urllib.request.urlopen(f"http://{SERVER}/history/{prompt_id}").read())
img = next(iter(hist[prompt_id]["outputs"].values()))["images"][0]
q   = urllib.parse.urlencode(img)
print("done:", f"http://{SERVER}/view?{q}")