ComfyUI-RefSampling

This is a proof-of-concept repo that implements:

• "Reference CNet" found in popular Image generation applications such as Automatic11
  • Allows image generation that adheres content to a reference image
• Visual Style Prompting
  • Allows image generation that adheres style to a reference image

Both of these techniques utilize attention injection from reference images into the sampled/generated image.

Examples

Text to Image Generation

Image Prompt: "orange gorilla stripes"

Visual Style Prompt: "black and orange stripes"

Image to Image Generation

Image Prompt: "an old man"

Visual Style Prompt "blue dragon, flames"