ComfyUI-Inspire-Pack - Regional IPAdapter

In this video, I'll be introducing a convenient feature of the recently added Attention Mask of ComfyUI_IPAdapter_Plus through the Inspire Pack. This feature allows easy utilization of the Attention Mask added not long ago. The Attention Mask is used internally to designate areas during a step called Attention. NOTE:The nodes provided by the "Inspire Pack" offer convenient features, and the ComfyUI_IPAdapter_Plus node must be installed for them to function. The "Regional IPAdapter..." node in the Inspire Pack takes inputs of image or embed and mask, helping apply IPAdapter to specific regions. The usage is similar to using "Regional Conditioning" and "Regional Sampler" from before. When using Color Mask, you can set the areas where IPAdapter will be applied through the "Regional IPAdapter By Color Mask" node. If you prefer to use Mask directly instead of Color Mask, you can use the ".. Mask" node. Let's divide the image into three regions for two reference characters and a background. If you want to use embeds instead of an image, use the "Regional IPAdapter Encoded..." node. Set the background to BLUE(#0000FF), and the two characters to RED(#FF0000) and GREEN(#00FF00) respectively. Generate a blue background image using EmptyImage. Then, draw the red and green regions on top of it. The ImageRefiner was used here as an example for drawing, but feel free to use any method you prefer. The configured IPAdapter for each region is unified into one through the "Apply Regional IPAdapters" node and is reflected in the MODEL. Connecting the MODEL output to KSampler generates result images with Attention Masks applied to each region. You can see that the properties of the Reference Image are reflected in each region. This allows for convenient utilization of techniques like "Instant LoRA".

Channel: Dr.Lt.Data's ComfyUI Extension

Views: 3,702

Keywords: ComfyUI, Impact Pack, Workflow Component, Stable Diffusion, Custom Nodes, AI

Id: EH2ojOWgUe8

Length: 6min 7sec (367 seconds)

Published: Wed Nov 22 2023