How do I make the result follow the edges more or less strictly?

Increase ControlNetApplyAdvanced weight (e.g., 0.8–1.0) for strong adherence to edges; decrease it (e.g., 0.3–0.6) for looser, more creative outputs. You can also shorten the guidance window (raise start or lower end) to reduce influence during part of the sampling.

What Canny thresholds should I use?

Start with a moderate range (e.g., low 80–120, high 160–220). Lower values capture more fine detail but can introduce noise; higher values simplify to bold outlines. Always preview the Canny output and aim for clean, unbroken primary contours.

Do I have to match the edge map size to the output size?

Yes—consistent results come from matching resolutions. Use ImageScale to resize the Canny edge map to the same width/height as EmptySD3LatentImage so ControlNet conditioning aligns with the sampling canvas.

How do I use a negative prompt with this workflow?

By default, ConditioningZeroOut supplies an empty negative. To add one, create another CLIPTextEncode with your negative text (e.g., "blurry, low quality"), and connect its output to KSampler’s negative conditioning input.

SD3.5 Large Canny ControlNet - ComfyUI Workflow

This ComfyUI workflow uses Stability AI’s SD 3.5 Large model with a Canny ControlNet to generate images that follow the structure of an input image’s edges. The pipeline loads the SD 3.5 checkpoint (CheckpointLoaderSimple) and the Canny ControlNet (ControlNetLoader), encodes your text prompt (CLIPTextEncode), and extracts a high-contrast edge map from your guide image (Canny). The edge map is applied as a structural constraint via ControlNetApplyAdvanced so the KSampler can synthesize new content that respects the outlines while still following your prompt.

Technically, the workflow creates a latent canvas at your target resolution (EmptySD3LatentImage), conditions the model with your positive prompt plus the ControlNet edge guidance, and optionally zeroes the negative prompt (ConditioningZeroOut) for simplicity. After sampling (KSampler), the result is decoded to RGB (VAEDecode), previewed (PreviewImage), and saved (SaveImage). ImageScale is included to match the Canny edge map to the chosen output size for consistent alignment. This setup is ideal when you want strong compositional control—like sticking to silhouettes or product contours—while letting SD 3.5 handle detail, texture, and style.

API

Use this workflow from code

Every Comfy workflow is a JSON graph. The payload below is this workflow, exactly as ComfyUI runs it — fetch it from the URL, keep it in version control, or load it in ComfyUI and run it node by node.

GET https://comfy.org/workflows/download/0bb057fd76e3.json

Fetching workflow JSON…

Run it from Python or TypeScript with the Comfy SDK. The same code targets Comfy Cloud or a ComfyUI you host yourself — only the base URL changes.

# Install (beta)
pip install comfy-sdk        # Python
npm i @comfyorg/sdk          # TypeScript

# Run "SD3.5 Large Canny ControlNet" (Python)
from comfy_sdk import Comfy

client = Comfy(api_key="comfyui-...")

# This workflow, exported in API format (see note below)
wf = client.workflows.from_file("sd3.5_large_canny_controlnet_example_api.json")

asset = client.assets.from_file("input.png")
wf.set_input("45", "image", asset)  # LoadImage

wf.set_input("6", "text", "crystal pink dragon on a blue mystery sky, hyperdetailed")  # CLIPTextEncode

job = client.run(wf)
for output in job.get_outputs("9"):  # SaveImage
    output.to_file(output.name)

The SDK takes a workflow in API format: open this workflow in ComfyUI and use File → Export Workflow (API). API access requires a Comfy Cloud plan with an API key. SDK docs · Get an API key

FAQ

Frequently Asked Questions

View all workflows