SD3.5 Large Canny ControlNet
Generate images guided by edge detection using SD 3.5 Canny ControlNet.
About
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.
What you provide
- Reading the workflow
Runs in ComfyUI. This workflow can't run on this page. Open it in Comfy Cloud with everything loaded, or download it for your own ComfyUI.
Try in CloudQuestions & answers
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.
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.
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.
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.