# SliderEdit Gives Fine-Grained, Continuous Control Over Instruction-Based Image Editing

_Research · published 2026-06-17_

Researchers from the University of Maryland and Adobe have introduced SliderEdit, a framework that adds continuous, per-instruction intensity control to existing instruction-based image editing models. Accepted as an oral presentation at CVPR 2026, the method trains a single set of lightweight low-rank adaptation matrices — rather than separate fine-tunes per attribute — that can suppress or modulate individual instructions within a multi-part prompt, exposing each as an adjustable slider.

SliderEdit has been applied to state-of-the-art models including FLUX-Kontext and Qwen-Image-Edit, where the authors report improvements in edit controllability, visual consistency, and user steerability. The core training objective, called Partial Prompt Suppression (PPS), teaches the adapter to neutralize the denoising influence of a specific instruction while leaving all other edits intact, enabling smooth interpolation along individual edit dimensions without disrupting spatial or semantic coherence.

## Sources
- [armanzarei.github.io](https://armanzarei.github.io/SliderEdit/)
---
Canonical: https://genbuzz.news/posts/slideredit-gives-fine-grained-continuous-control-over-instruction-based-image-editing