Important Nodes to Use

The Workflow Editor has a large number of nodes designed for specific use cases, but there are a few key nodes will be used most often.

Denoising Latents & Noise

The 'Denoise Latents' node is a central node in the Workflow Editor. The node takes many different inputs including the positive & negative conditioning, initial random noise, as well as any ControlNets, IP-Adapter or masks that might be used. 

Generally, the 'Denoise Latents' node will be connected to a 'Noise' node. This 'Noise' node creates a noised image of the selected width and height. To use a random seed to created the noised image, a 'Random Integer' node can be connected the 'Noise' node. This is shown in the image below:

Model Loading, Prompts & Conditioning

Model loading is done through the 'Main Model' node. The CLIP will usually connect to the Prompt nodes, with the UNet being passed to the 'Denoise Latents' node.

Conditioning is necessary for the latent diffusion process, whether empty or not. As a result, the denoising node requires positive and negative conditioning inputs. Conditioning is reliant on a CLIP text encoder provided by the 'Model Loader' node. This is show in image below: 

ControlNets

The 'ControlNet' node outputs a control, which can be provided as input to a 'Denoise Latents' node. Depending on the type of ControlNet desired, ControlNet nodes might require an image processor node, such as a 'Canny', 'Depth' or 'OpenPose Processor', which prepares an input image for use with ControlNet. Multiple ControlNets can be used together by passing the ControlNets into a 'Collection' node and passing the collection to the 'Denoise Latents' node, as show below:

LoRAs

The 'Lora Loader' node lets you load a LoRA and pass it as output. The LoRA node is used similarly to the 'Main Model Loader' node, but sits in-between the 'Prompt' nodes and the 'Main Model' node.