SliderEdit: Continuous Image Editing with Fine-Grained Instruction Control
Arman Zarei, Samyadeep Basu, Mobina Pournemat, Sayan Nag, Ryan Rossi, Soheil Feizi
2025-11-14
Summary
This paper introduces a new way to edit images using instructions, giving users much more precise control over *how* those instructions are applied.
What's the problem?
Current image editing models that use instructions treat all parts of an instruction equally, applying them with the same strength. This is limiting because users can't easily fine-tune individual changes or adjust how much of each instruction is actually followed. Imagine asking a model to 'make the sky bluer and add a sunset' – you can't easily say 'make the sky *slightly* bluer, but a *very* vibrant sunset'.
What's the solution?
The researchers created a system called SliderEdit. It breaks down each instruction into separate 'sliders' that control the strength of that specific change. So, for the sky and sunset example, you'd have a 'blueness' slider and a 'sunset vibrancy' slider. Importantly, this system doesn't need to be retrained for every new type of edit; it learns a general way to adjust edits across many different images and instructions using a clever technique with 'low-rank adaptation matrices'. This allows for smooth, continuous changes while keeping the image looking realistic and consistent.
Why it matters?
This work is important because it's the first to offer this level of detailed, continuous control within instruction-based image editing. It moves beyond simply telling an image what to change to letting users precisely *how much* to change it, opening the door for more interactive and creative image manipulation.
Abstract
Instruction-based image editing models have recently achieved impressive performance, enabling complex edits to an input image from a multi-instruction prompt. However, these models apply each instruction in the prompt with a fixed strength, limiting the user's ability to precisely and continuously control the intensity of individual edits. We introduce SliderEdit, a framework for continuous image editing with fine-grained, interpretable instruction control. Given a multi-part edit instruction, SliderEdit disentangles the individual instructions and exposes each as a globally trained slider, allowing smooth adjustment of its strength. Unlike prior works that introduced slider-based attribute controls in text-to-image generation, typically requiring separate training or fine-tuning for each attribute or concept, our method learns a single set of low-rank adaptation matrices that generalize across diverse edits, attributes, and compositional instructions. This enables continuous interpolation along individual edit dimensions while preserving both spatial locality and global semantic consistency. We apply SliderEdit to state-of-the-art image editing models, including FLUX-Kontext and Qwen-Image-Edit, and observe substantial improvements in edit controllability, visual consistency, and user steerability. To the best of our knowledge, we are the first to explore and propose a framework for continuous, fine-grained instruction control in instruction-based image editing models. Our results pave the way for interactive, instruction-driven image manipulation with continuous and compositional control.