SliderEdit++: Improved Slider Trajectories in Instruction-Based Image Editing

Key points

• Slider-based control aims to give users continuous, fine-grained control over the intensity of individual edits in instruction-based image editing.1
• However, existing approaches fail to deliver reliable continuous controllability: traversing a slider can produce abrupt visual jumps and non-monotonic changes, while jointly controlled sliders can interfere with one another.1
• Building on these observations, we propose SliderEdit++, which explicitly regularizes the intermediate editing trajectory rather than supervising only its endpoints.1
• Our method combines velocity- and semantic-space trajectory regularization with spectral control to encourage smooth, predictable slider traversal, together with a disentanglement objective that reduces interference between independently controlled edits.1

Sources (1)

• [1]SliderEdit++: Improved Slider Trajectories in Instruction-Based Image Editing
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 07:14 PM

  To systematically study these limitations, we introduce SlideBench, a comprehensive benchmark for continuous image editing spanning diverse editing categories and equipped with a unified suite of metrics quantifying monotonic progression, perceptual continuity, and inter-slider disentanglement.

  Slider-based control aims to give users continuous, fine-grained control over the intensity of individual edits in instruction-based image editing.

Extractive summary: sentences quoted from the sources.