Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces FlowSlider, a training-free approach for continuous image editing within the Rectified Flow framework. FlowSlider decomposes the image update into orthogonal fidelity and steering terms, allowing for manipulation of edit strength by scaling only the steering component. Experiments demonstrate that FlowSlider achieves smooth and reliable control over edit strength without requiring any post-training, outperforming existing methods across various image editing tasks.
Ditch the training wheels: FlowSlider unlocks continuous image editing with unprecedented control and fidelity, all without needing any additional training.
Continuous image editing aims to provide slider-style control of edit strength while preserving source-image fidelity and maintaining a consistent edit direction. Existing learning-based slider methods typically rely on auxiliary modules trained with synthetic or proxy supervision. This introduces additional training overhead and couples slider behavior to the training distribution, which can reduce reliability under distribution shifts in edits or domains. We propose \textit{FlowSlider}, a training-free method for continuous editing in Rectified Flow that requires no post-training. \textit{FlowSlider} decomposes FlowEdit's update into (i) a fidelity term, which acts as a source-conditioned stabilizer that preserves identity and structure, and (ii) a steering term that drives semantic transition toward the target edit. Geometric analysis and empirical measurements show that these terms are approximately orthogonal, enabling stable strength control by scaling only the steering term while keeping the fidelity term unchanged. As a result, \textit{FlowSlider} provides smooth and reliable control without post-training, improving continuous editing quality across diverse tasks.