UniSlider: Perceptually Uniform Sliders for Continuous Image Editing
UniSlider separates the visible edit pace from the model’s raw strength setting.
The paper says common image-editing sliders move unevenly: some ranges barely change the image, while others jump or partly reverse. UniSlider trains a lightweight LoRA so perceptual distance from the input grows more linearly as the slider moves. An inference-time remapping then closes the remaining gap without extra training or parameters. The authors report stronger results than prior methods on a 300-edit benchmark and in a user study. ArXiv · AI/CL/LG's note
The paper says common image-editing sliders move unevenly: some ranges barely change the image, while others jump or partly reverse. UniSlider trains a lightweight LoRA so perceptual distance from the input grows more linearly as the slider moves. An inference-time remapping then closes the remaining gap without extra training or parameters. The authors report stronger results than prior methods on a 300-edit benchmark and in a user study. ArXiv · AI/CL/LG's note
score 4