Stable Diffusion ComfyUi Control Net Deep Dive

This Video takes a deep dive into Stable Diffusion + ComfyUi + ControNet. It looks at Canny Edge Detection, Image Depth, and Scribble / Sketch.

ArchitectureRealMix Checkpoint:

Connect with me on Instagram

#stabledifussion #comfyui #controlnet

[hatonafox]

You don't need to uninstall them everytime you close it. You just need to make sure you don't use them in the same workflow that you use any of the possible conflicting ones. In general working mixing control net custom nodes from different developers isn't a good idea the reason being that the nodes are expecting very specific output and input values. Some play well together but most don't.

For the images you load into the control net with certain images you might want to resize your image using a resizing node prior to loading it into the pre-processor to ensure you maintain proportions while working within an ideal image resolution.

Lastly, the prompting style you're using is much better suited to SDXL which handles natural language better because of how the CLIP model is trained. You are correct that the CLIP model training can play a part in how elements of an image are generated but in general these use SD1.5 or SDXL as a base so as long as the prompting works well in base for retrieving elements (e.g. a crowd of people) it will work for the tuned models. There are exceptions.