CanFuUI: A Canvas-Centric Web User Interface for Iterative Image Generation with Diffusion Models and ControlNet
摘要
Today, various AI generation tools are emerging in succession. And the majority of existing tools are predominantly model-centric in design, resulting in steep learning curves and high usability thresholds for users. Moreover, current user interfaces lack built-in image editing capabilities, forcing users to rely on external software even for basic image editing tasks. Considering that most image generation is an iterative process, this limitation significantly hampers user experience and creative potential. Instead, this paper proposes a novel canvas-centric design that seamlessly integrates editing functionalities into the UI called CanFuUI, streamlining secondary image processing. Users can crop, modify, and annotation of specific regions of generated images within the same canvas in CanFuUI. Furthermore, canvas content is utilized as preprocessed images, directly integrated into the ControlNet preprocessing procedure, reinforcing the customization capabilities of AI-generated outputs.